Pith. sign in

REVIEW 3 major objections 4 minor 239 references

SAFE-Merge: Data-Free Continual Model Merging with General Knowledge Preservation

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read SAFE-Merge protects a merged model's general knowledge by masking risky parameter updates, then recovers lost task skill with a low-rank correction.

desk verdict A well-evidenced continual-merging method that deserves review, but its headline advantage may depend on hyperparameters chosen on the evaluation tasks themselves. read the letter →

arxiv 2608.01184 v1 pith:YWBGMCXV submitted 2026-08-02 cs.LG

Zihuan Qiu , Zhiyang Liao , Chiyuan He , Yi Xu , Fanman Meng , Linfeng Xu , Qingbo Wu , Hongliang Li This is my paper
classification cs.LG
keywords continualmodelmergingdata-freegeneralknowledgepreservationrisk-awaremaskinglow-rankrecoverytaskvectorsH-scoreCLIP
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper takes on a blind spot in data-free continual model merging: existing methods protect previously merged tasks from interference but treat the pretrained backbone as a neutral container, so general knowledge silently erodes. SAFE-Merge claims that this erosion can be prevented without any task data by deciding, parameter by parameter, which updates are safe. It scores each entry of an incoming task vector by contrasting its energy along directions of the pretrained weights (risky) against directions unique to the new task (valuable), masks out the high-risk entries, and then recovers most of the lost task information with a low-rank correction confined to the kept entries. On CLIP vision streams of 8, 14, and 20 tasks and on five GLUE tasks with Flan-T5, it reports the best harmonic mean of merged-task accuracy and held-out general performance, with the gap over baselines growing as sequences lengthen. The payoff, if right, is that a deployment system can keep folding in specialized checkpoints without sacrificing the model's broader competence.

What carries the argument

The method has two coupled parts. (1) Risk-aware sparse masking: for each linear layer, compute $Q_0^{(l)}$ from the top-$r_p$ right singular vectors of the pretrained weight and $Q_{\mathrm{new}}^{(l)}$ from the incoming task vector's right singular vectors after removing its overlap with $Q_0^{(l)}$; the per-entry score $R^{(l)}_{ij} = (\tau_t^{(l)} Q_0^{(l)}{Q_0^{(l)}}^\top)^2_{ij} - (\tau_t^{(l)} Q_{\mathrm{new}}^{(l)}{Q_{\mathrm{new}}^{(l)}}^\top)^2_{ij}$ separates overwrite risk from task utility, and the mask $M_t$ keeps the lowest-scoring $\rho$ fraction. (2) Masked low-rank recovery: a low-rank factor pair $(B_t^{(l)}, A_t^{(l)})$ is optimized, its product masked by $M_t$, so no rej

What would settle it

On a backbone where the pretrained weights' singular directions are known not to align with the general data covariance (e.g., a CLIP text tower applied to a vision stream, or a randomly permuted weight matrix), run the positive-versus-negative selection experiment from the paper's Figure 1: if retaining low-risk entries no longer keeps held-out accuracy near the pretrained level while high-risk selection does no worse, the proxy fails.

Watch

Extended reading notes

Core claim

The central claim is that parameter-wise overwrite risk can be read off parameters alone. Treating the pretrained weight matrix as a generalized update, its top-$r_p$ right singular vectors $V_0^{(l)}$ approximate the dominant input-activation subspace of the general training distribution; the risk score $R^{(l)}_{ij}$ in Eq. (6) then quantifies, for each entry of the incoming task vector $\tau_t^{(l)}$, whether its projected energy falls mostly on those general directions or on the orthogonalized task-unique directions $Q_{\mathrm{new}}^{(l)}$. Retaining only the low-risk entries (the mask $M_t$) and adding a low-rank recovery $M_t \odot (\tau_t + B_t A_t)$ optimized against old-task and ne

Load-bearing premise

The load-bearing premise is that the top right singular vectors of the pretrained weight matrix capture the same directions that carry large energy in the general training distribution's input covariance; if that correspondence breaks on a different backbone or under distribution shift, the mask will protect the wrong parameters and the general-knowledge guarantee is lost.

Editorial extensions

If this is right

  • If correct, data-free continual merging can protect held-out generalization without any calibration data or replay, since the risk score is computed purely from pretrained, merged, and incoming parameters.
  • The safety-rather-than-interference framing means the benefits should grow as more tasks accumulate, and indeed the paper reports the largest H-score gains on the 20-task stream.
  • The fused update adds no inference-time parameters or latency, so the preservation comes for free at deployment.
  • The positive/negative selection experiment offers a reusable diagnostic: the risk score can rank any task vector's entries by how much they threaten general knowledge.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the singular-direction proxy transfers beyond CLIP and Flan-T5, the same score could act as a general safety filter for deciding which user-adapted checkpoints are safe to aggregate or share.
  • The keep ratio $\rho$ is scheduled by backbone and stream length; a natural extension is to make it adaptive per layer or per incoming task using the separation between the risk-score distributions.
  • The mask-plus-recovery split suggests a general design: spend a hard safety budget on a small, trusted set of coordinates and use a constrained correction to squeeze task value out of that same set — an idea that could apply to other plasticity-stability settings.
  • A testable consequence: apply the risk score to partially fine-tuned or LoRA-adapted models; if the weight-to-covariance correspondence holds there, the method extends beyond full fine-tuning.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. SAFE-Merge proposes a data-free continual model merging method that first masks incoming task-vector updates according to a per-parameter risk score, then learns a masked low-rank recovery term, and finally fuses the result into the running backbone. The risk score contrasts energy along right singular vectors of the pretrained weights (a proxy for general-knowledge-sensitive directions) with energy along task-specific directions. The paper reports that SAFE-Merge consistently achieves the best harmonic mean of merged-task accuracy and held-out general performance across CLIP vision streams of 8, 14, and 20 tasks, three CLIP backbones, ten task orders, and a Flan-T5 GLUE benchmark, while adding no inference-time parameters. An ablation study and rank-sensitivity analysis support the design choices, and a positive-versus-negative selection experiment (Fig. 1) gives direct evidence that the risk score orders parameter updates by their damage to held-out general accuracy.

Significance. If the central claims hold, SAFE-Merge is a worthwhile step for data-free continual merging: it explicitly targets preservation of pretrained general knowledge rather than only downstream-task interference, and it does so with a simple parameter-space procedure that leaves masked parameters untouched and adds no inference cost. The empirical coverage is unusually thorough for this area: three backbones, three stream lengths, ten task orders, ablations, rank sensitivity, a language benchmark, and an overhead table. The positive/negative selection experiment in Fig. 1 is a particularly good falsifiable check of the risk-scoring mechanism. However, the paper's central claim of a "data-free" advantage is weakened by the use of the evaluation benchmarks themselves to select key hyperparameters, and the principal risk-scoring proxy is validated on only one backbone. Neither issue is fatal, but both need to be addressed before the claims can be accepted at face value.

major comments (3)
  1. [§Experiments/Implementation Details, Fig. 3, Appendix A.6] The keep ratio ρ and plasticity weight λ are selected using the evaluation benchmarks themselves. Fig. 3(a) plots ACC/Gen. on the exact 8/14/20-task streams with the exact IN/IN-R/ObjNet probes for the ρ values later used in Tables 1 and S.5, and Fig. 3(b) does the same for λ; A.6 states the knees “motivate stronger sparsification for longer sequences,” so the final ρ values are read off these curves. No held-out split or nested protocol is described. Since the baselines use their original, not per-benchmark, hyperparameters, the reported “data-free” gains and the claim of consistently best H-score may be an artifact of oracle tuning. Please add a validation protocol (e.g., tune on one held-out task set and evaluate on another, or fix ρ/λ a priori) and show the gains survive.
  2. [§Motivation, Eq. (3), Appendix A.1] The risk score replaces the unavailable general-data covariance C_g with Q_0 Q_0^T, the top right singular subspace of the pretrained weight. Appendix A.1 explicitly calls this a proxy, not an exact identity, and Fig. 1 validates it on ViT-B/32 only. The guarantee that low-risk masks protect general knowledge depends on this correspondence holding for the deployment distribution. Please provide the same positive/negative validation for ViT-B/16 and ViT-L/14 (and ideally the NLP backbone), or an explicit comparison with a data-derived covariance on a held-out general set; otherwise the central risk-ordering claim is under-supported.
  3. [§Appendix A.2, Eq. (S.3)–(S.5)] The theoretical justification of the recovery objective imports Corollary 1 of Qiu et al. (2026) without proof and without checking the misalignment term ζ or the data-dependent scales σ_1(X). The paper correctly states that the result only motivates the objective, but the main-text presentation nevertheless frames the recovery as solving the preservation problem. The gap between an unmeasured upper bound and the actual loss should be stated more prominently in the main text, or the corollary should be proved in the appendix under the specific SAFE-Merge variables.
minor comments (4)
  1. [Table 3] The Naive accumulation and Risk-aware mask-only rows report ±0.0 even though results are said to be averaged over ten task orders; task order should affect these quantities. Please clarify whether these rows use a single order or whether the zeros are a typographical artifact.
  2. [§Experiments/Implementation Details] The phrase “without task-specific tuning” is contradicted by ρ varying with stream length and backbone (0.5/0.3 for ViT-B/32, 0.7/0.5 for ViT-B/16 and L/14). Rephrase to say ranks are shared while ρ is adjusted by protocol.
  3. [§Experiments, reproducibility] No code or checkpoint release is indicated. For reproducibility, please release the merge script, the exact GF/FusionBench checkpoint identifiers, and the task permutations used (Table S.1 is helpful but not machine-readable).
  4. [Fig. 1 caption] The terms “positive selection” and “negative selection” are defined only in the running text; add a one-line definition in the caption so the figure stands alone.

Circularity Check

2 steps flagged · score 4.0 of 10

H-score gains are partly in-sample: ρ and λ are chosen from the same benchmarks whose H-score is then reported; recovery derivation also leans on a same-author corollary.

  1. fitted input called prediction [Ablation and Analysis Results (Fig. 3) + Appendix A.6 + Implementation Details]
    "The knees at ρ= 0.3 –0.5 motivate stronger sparsification for longer sequences. ... We choose 0.5 and 0.3 for ViT-B/32 under the 8-task and 14/20-task protocols, respectively."

    Fig. 3(a) plots ACC–Gen for the exact 8/14/20-task streams and the IN/IN-R/ObjectNet probes that later appear in Table 1; Fig. 3(b) plots ACC–BWT for the same streams. The chosen ρ and λ are read off these curves, and then the same curves' benchmarks are reported as the method's H-score. The reported 'consistently best H-score' is therefore not an independent prediction from a fixed data-free configuration but an in-sample selection of the sparsity and plasticity hyperparameters on the evaluation tasks themselves. A deployer without access to these exact benchmarks could not reproduce the configuration, so the data-free advantage claimed in the abstract is partially an artifact of benchmark tuning.

  2. self citation load bearing [Appendix A.2, Data-Free Derivation of the Recovery Objectives]
    "Corollary 1 of Qiu et al. (2026) directly gives the data-free inequality used here."

    The recovery objectives Eq. (10)–(12) are claimed to be derived from data-dependent losses, but the bridge to the data-free form is an inequality imported from the authors' own prior NUFILT paper (Qiu et al. 2026), which heavily overlaps in authorship. The corollary is not re-derived or machine-checked in this manuscript, so the 'first-principles' justification of the recovery term reduces to a load-bearing self-citation. This does not by itself force the empirical H-score result—Table 3's ablation independently shows recovery helps—but it makes the theoretical derivation of a central component depend on the same group's prior unverified claim.

full rationale

The core derivation chain is not circular in the strict equation-identity sense: the risk score in Eq. (4) is a defined proxy based on pretrained-weight singular directions, and its validation in Fig. 1 is a controlled positive/negative experiment; the recovery losses in Eq. (10)–(11) are objectives, not predictions. No equation in the paper is equivalent by construction to the reported H-score. However, the headline claim of 'consistently best H-score' is weakened by the fact that the key hyperparameters ρ and λ (and the ranks) are selected from curves measured on the exact benchmarks and held-out probes that later produce the reported tables. This is an in-sample tuning of the configuration rather than a fixed data-free protocol, so the comparison against baselines with their original hyperparameters is not fully fair. Additionally, the data-free derivation of the recovery objective imports Corollary 1 from the authors' prior same-group paper without independent verification, which is a load-bearing self-citation for the theoretical justification though not for the empirical outcome. Overall, these issues constitute partial, not complete, circularity.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

SAFE-Merge introduces no new physical entities or latent variables beyond parameter-derived subspaces and a scalar risk score, which are operational proxies rather than invented objects. The main hidden costs are the hyperparameters listed above and the load-bearing proxy assumptions connecting pretrained weight singular vectors to general-data covariance. The heaviest dependency is the self-cited upper bound from Qiu et al. (2026), which carries part of the theoretical justification for the recovery objective.

free parameters (6)
  • keep ratio rho = 0.5 (ViT-B/32, 8 tasks); 0.3 (ViT-B/32, 14/20); 0.7 (ViT-B/16, L/14, 8); 0.5 (ViT-B/16, L/14, 14/20); 0.9 (Flan-T5)
    Sets the masking quantile threshold. Selected from the ACC-Gen trade-off curves in Fig. 3a and per-backbone/stream choices, so benchmark outcomes drove the selection.
  • plasticity weight lambda = 0.8
    Chosen from the ACC-BWT sensitivity analysis in Fig. 3b; balances task recovery against preservation but is tuned on benchmark metrics.
  • scoring rank rp = 128
    Selected from rank sensitivity in Fig. 4a; larger ranks degrade ACC and BWT.
  • task-subspace rank rv = 8
    Selected from rank sensitivity in Fig. 4b; rv=8 gives best ACC, while rv=4 gives least-negative BWT.
  • recovery rank rl = 64
    Stable across 16-128 in Fig. 4c; selected from the broad stable region.
  • regularization weight mu = 0.1
    Fixed throughout; bounds the trainable part of spectral residual terms in the recovery derivation, but is not ablated in the main paper.
assumptions (7)
  • domain assumption All checkpoints share the same pretrained initialization and architecture.
    Stated in Limitations (Appendix C.1). Required so task vectors and parameter singular directions are comparable.
  • domain assumption The pretrained model is close enough to a stationary point that the first-order term in Eq. (1) vanishes.
    Used to replace the gradient of the general-knowledge loss by curvature. This is an idealization for a pretrained model rather than an exact property.
  • ad hoc to paper Top right singular vectors of the pretrained weight approximate the dominant input activation covariance subspace of general data.
    Eq. (3) substitutes C_hat_g = Q0 Q0^T for C_g. Motivated by an observation from Qiu et al. (2026), validated only in Fig. 1, and admitted as a proxy in Appendix A.1/A.2.
  • ad hoc to paper Corollary 1 of Qiu et al. (2026) bounds the data-dependent response losses and applies to SAFE-Merge recovery variables.
    Appendix A.2 derives the recovery objectives Eq. (10)/(11) from this self-cited bound; it is not re-proved or machine-checked here.
  • domain assumption Incoming task checkpoints lie in the same low-loss basin or aligned mode as the pretrained model, so additive task vectors are meaningful.
    Inherited from the task arithmetic and model merging literature and used throughout the merging process.
  • domain assumption Response change in linear layers proxies general-knowledge preservation.
    Eq. (2) defines the squared response-change proxy; Appendix A.1 explicitly says this is a surrogate, not an exact identity, for the loss Hessian.
  • ad hoc to paper The accumulated merged vector tau_bar_{t-1} is a valid data-free proxy for earlier-task activation directions.
    Eq. (9) uses V_old from tau_bar_{t-1} to build the forgetting penalty; no data is available to verify the proxy.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SAFE-Merge: Data-Free Continual Model Merging with General Knowledge Preservation." pith.science (2026). https://pith.science/paper/YWBGMCXV

@misc{pith2026260801184,
  author       = {Pith},
  title        = {Pith review of: SAFE-Merge: Data-Free Continual Model Merging with General Knowledge Preservation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YWBGMCXV}},
  note         = {Machine review of arXiv:2608.01184}
}
read the original abstract

Data-free continual model merging must incorporate a stream of specialized models while retaining both pretrained general knowledge and previously acquired tasks, without access to task data. Existing methods mainly merge task updates by suppressing interference among downstream tasks; while this protects previously acquired tasks, it overlooks the safety of the pretrained knowledge itself, whose erosion degrades generalization to held-out distributions and weakens the foundation for future task acquisition. We propose SAFE-Merge, a simple data-free continual-merging framework that first decides which parameter updates are safe to retain, and then recovers the task information lost through masking. Specifically, to ensure safety, risk-aware sparse masking selects parameter updates that carry task-specific information while posing low risk to general knowledge. Masked low-rank recovery then compensates for the lost task information using only the same retained parameter updates, while leaving all masked-out parameters strictly unchanged. Finally, the combined update is fused into the backbone, incurring no additional inference cost. Across vision and language benchmarks, SAFE-Merge consistently achieves the best H-score. On longer CLIP task sequences, it substantially improves H-score over NUFILT while also achieving the highest accuracy.

Figures

Figures reproduced from arXiv: 2608.01184 by the authors.

Figure 1
Figure 1. confirms the score. We merge 8 tasks sequentially by adding only the masked portion of each incoming task vector to the running merged model. Retaining the lowest-risk parameter updates produces a broad flat region: performance stays close to the pretrained model on all three datasets even when up to 50% of each task vector is kept, while reversing the order causes a large drop under the same budget. Since the two s… view at source ↗
Figure 2
Figure 2. Overview of SAFE-Merge. SAFE-Merge identifies low-risk parameter updates in the incoming task vector, recovers task information over the retained subset using a masked low-rank update, and fuses the resulting update into the running merged model without retaining extra inference-time parameters. preservation cost against task utility: R (l) ij =  τ (l) t Q (l) 0 Q (l)⊤ 0 2 ij −  τ (l) t Q (l) newQ (l)⊤ new 2 ij … view at source ↗
Figure 3
Figure 3. Analysis of risk-aware masking and stability– [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Rank sensitivity on CLIP ViT-B/32 under the 8-task continual merging protocol. We report ACC and BWT while [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

239 extracted references · 37 canonical work pages

  1. [1]

    International conference on machine learning , pages=

    Test-time training with self-supervision for generalization under distribution shifts , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  2. [2]

    arXiv preprint arXiv:2006.10726 , year=

    Tent: Fully test-time adaptation by entropy minimization , author=. arXiv preprint arXiv:2006.10726 , year=

    arXiv 2006

  3. [3]

    International Journal of Computer Vision , volume=

    A comprehensive survey on test-time adaptation under distribution shifts , author=. International Journal of Computer Vision , volume=. 2025 , publisher=

    2025

  4. [4]

    International conference on machine learning , pages=

    Efficient test-time model adaptation without forgetting , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  5. [5]

    Advances in Neural Information Processing Systems , volume=

    Test-time classifier adjustment module for model-agnostic domain generalization , author=. Advances in Neural Information Processing Systems , volume=

  6. [6]

    arXiv preprint arXiv:2210.11466 , year=

    Surgical fine-tuning improves adaptation to distribution shifts , author=. arXiv preprint arXiv:2210.11466 , year=

    arXiv

  7. [7]

    International conference on machine learning , pages=

    Adanpc: Exploring non-parametric classifier for test-time adaptation , author=. International conference on machine learning , pages=. 2023 , organization=

    2023

  8. [8]

    Advances in neural information processing systems , volume=

    Improving robustness against common corruptions by covariate shift adaptation , author=. Advances in neural information processing systems , volume=

Show all 239 references
  1. [9]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models , author=. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

  2. [10]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Back to the source: Diffusion-driven adaptation to test-time corruption , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  3. [11]

    The Eleventh International Conference on Learning Representations , year=

    Towards Stable Test-time Adaptation in Dynamic Wild World , author=. The Eleventh International Conference on Learning Representations , year=

  4. [12]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Revisiting Flatness-aware Optimization in Continual Learning with Orthogonal Gradient Projection , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  5. [13]

    Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

    Training networks in null space of feature covariance for continual learning , author=. Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

  6. [14]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Continual test-time domain adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  7. [15]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Robust mean teacher for continual and gradual test-time adaptation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  8. [16]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Robust test-time adaptation in dynamic scenarios , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  9. [17]

    Advances in neural information processing systems , volume=

    Memo: Test time robustness via adaptation and augmentation , author=. Advances in neural information processing systems , volume=

  10. [18]

    arXiv preprint arXiv:2006.10963 , year=

    Evaluating prediction-time batch normalization for robustness under covariate shift , author=. arXiv preprint arXiv:2006.10963 , year=

    2006 arXiv

  11. [19]

    Advances in Neural Information Processing Systems , volume=

    Test-time prompt tuning for zero-shot generalization in vision-language models , author=. Advances in Neural Information Processing Systems , volume=

  12. [20]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Diverse data augmentation with diffusions for effective test-time prompt tuning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  13. [21]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Efficient test-time adaptation of vision-language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  14. [22]

    International conference on machine learning , pages=

    Do we really need to access the source data? source hypothesis transfer for unsupervised domain adaptation , author=. International conference on machine learning , pages=. 2020 , organization=

    2020

  15. [23]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Ecotta: Memory-efficient continual test-time adaptation via self-distilled regularization , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  16. [24]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Revisiting realistic test-time training: Sequential inference and adaptation by anchored clustering regularized self-training , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  17. [25]

    Youngjun Lee and Doyoung Kim and Junhyeok Kang and Jihwan Bang and Hwanjun Song and Jae-Gil Lee , booktitle=

  18. [26]

    The Thirteenth International Conference on Learning Representations , year=

    Noisy Test-Time Adaptation in Vision-Language Models , author=. The Thirteenth International Conference on Learning Representations , year=

  19. [27]

    Weight averaging for neural networks and local resampling schemes , author=. Proc. AAAI-96 Workshop on Integrating Multiple Learned Models. AAAI Press , pages=. 1996 , organization=

    1996

  20. [28]

    Proceedings of the 12th annual conference on Computer graphics and interactive techniques , pages=

    Animating rotation with quaternion curves , author=. Proceedings of the 12th annual conference on Computer graphics and interactive techniques , pages=

  21. [29]

    arXiv preprint arXiv:2110.06296 , year=

    The role of permutation invariance in linear mode connectivity of neural networks , author=. arXiv preprint arXiv:2110.06296 , year=

    arXiv

  22. [30]

    arXiv preprint arXiv:2209.04836 , year=

    Git re-basin: Merging models modulo permutation symmetries , author=. arXiv preprint arXiv:2209.04836 , year=

    arXiv

  23. [31]

    Advances in Neural Information Processing Systems , volume=

    Task arithmetic in the tangent space: Improved editing of pre-trained models , author=. Advances in Neural Information Processing Systems , volume=

  24. [32]

    The Eleventh International Conference on Learning Representations , year=

    Dataless Knowledge Fusion by Merging Weights of Language Models , author=. The Eleventh International Conference on Learning Representations , year=

  25. [33]

    The Twelfth International Conference on Learning Representations , year=

    Parameter-Efficient Multi-Task Model Fusion with Partial Linearization , author=. The Twelfth International Conference on Learning Representations , year=

  26. [34]

    The Twelfth International Conference on Learning Representations , year=

    AdaMerging: Adaptive Model Merging for Multi-Task Learning , author=. The Twelfth International Conference on Learning Representations , year=

  27. [35]

    Advances in Neural Information Processing Systems , volume=

    Ties-merging: Resolving interference when merging models , author=. Advances in Neural Information Processing Systems , volume=

  28. [36]

    Forty-first International Conference on Machine Learning , year=

    Language models are super mario: Absorbing abilities from homologous models as a free lunch , author=. Forty-first International Conference on Machine Learning , year=

  29. [37]

    arXiv preprint arXiv:2212.01378 , year=

    Cold fusion: Collaborative descent for distributed multitask finetuning , author=. arXiv preprint arXiv:2212.01378 , year=

    arXiv

  30. [38]

    Advances in Neural Information Processing Systems , volume=

    Merging models with fisher-weighted averaging , author=. Advances in Neural Information Processing Systems , volume=

  31. [39]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    LoRA-Flow: Dynamic LoRA Fusion for Large Language Models in Generative Tasks , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  32. [40]

    , author=

    Lora: Low-rank adaptation of large language models. , author=. ICLR , volume=

  33. [41]

    Findings of the Association for Computational Linguistics: EACL 2023 , pages=

    AdapterSoup: Weight Averaging to Improve Generalization of Pretrained Language Models , author=. Findings of the Association for Computational Linguistics: EACL 2023 , pages=

    2023

  34. [42]

    LoraHub: Efficient Cross-Task Generalization via Dynamic Lo

    Chengsong Huang and Qian Liu and Bill Yuchen Lin and Tianyu Pang and Chao Du and Min Lin , booktitle=. LoraHub: Efficient Cross-Task Generalization via Dynamic Lo

  35. [43]

    Findings of the Association for Computational Linguistics ACL 2024 , pages=

    LoraRetriever: Input-Aware LoRA Retrieval and Composition for Mixed Tasks in the Wild , author=. Findings of the Association for Computational Linguistics ACL 2024 , pages=

    2024

  36. [44]

    Merging Lo

    Ziyu Zhao and Tao Shen and Didi Zhu and Zexi Li and Jing Su and Xuwu Wang and Fei Wu , booktitle=. Merging Lo

  37. [45]

    Model merging with

    George Stoica and Pratik Ramesh and Boglarka Ecsedi and Leshem Choshen and Judy Hoffman , booktitle=. Model merging with

  38. [46]

    The Thirteenth International Conference on Learning Representations , year=

    Mitigating Parameter Interference in Model Merging via Sharpness-Aware Fine-Tuning , author=. The Thirteenth International Conference on Learning Representations , year=

  39. [47]

    International Conference on Machine Learning , pages=

    Merging Multi-Task Models via Weight-Ensembling Mixture of Experts , author=. International Conference on Machine Learning , pages=. 2024 , organization=

    2024

  40. [48]

    The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

    Merging on the Fly Without Retraining: A Sequential Approach to Scalable Continual Model Merging , author=. The Thirty-ninth Annual Conference on Neural Information Processing Systems , year=

  41. [49]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Boosting continual learning of vision-language models via mixture-of-experts adapters , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  42. [50]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Learning without forgetting for vision-language models , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  43. [51]

    European Conference on Computer Vision , pages=

    Class-incremental learning with clip: Adaptive representation adjustment and parameter fusion , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  44. [52]

    2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    Always Be Dreaming: A New Approach for Data-Free Class-Incremental Learning , author=. 2021 IEEE/CVF International Conference on Computer Vision (ICCV) , year=

    2021

  45. [53]

    Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

    On learning the geodesic path for incremental learning , author=. Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

  46. [54]

    Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

    Mnemonics training: Multi-class incremental learning without forgetting , author=. Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

  47. [55]

    IEEE Transactions on Knowledge and Data Engineering , volume=

    Learning adaptive embedding considering incremental class , author=. IEEE Transactions on Knowledge and Data Engineering , volume=. 2021 , publisher=

    2021

  48. [56]

    International conference on machine learning , pages=

    On the power of curriculum learning in training deep networks , author=. International conference on machine learning , pages=. 2019 , organization=

    2019

  49. [57]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Large scale incremental learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  50. [58]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Class-incremental learning by knowledge distillation with adaptive feature consolidation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  51. [59]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Riemannian walk for incremental learning: Understanding forgetting and intransigence , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  52. [60]

    In Advances in Neural Information Processing Systems (NeurIPS) Workshop , year=

    Distilling the knowledge in a neural network , author=. In Advances in Neural Information Processing Systems (NeurIPS) Workshop , year=

  53. [61]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Distance-Based Image Classification: Generalizing to New Classes at Near-Zero Cost , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  54. [62]

    2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Semantic Drift Compensation for Class-Incremental Learning , author=. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2020

  55. [63]

    Advances in Neural Information Processing Systems , volume=

    Understanding the role of training regimes in continual learning , author=. Advances in Neural Information Processing Systems , volume=

  56. [64]

    arXiv preprint arXiv:2006.10974 , year=

    Optimization and generalization of regularization-based continual learning: a loss approximation viewpoint , author=. arXiv preprint arXiv:2006.10974 , year=

    2006 arXiv

  57. [65]

    European Conference on Computer Vision , year=

    PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning , author=. European Conference on Computer Vision , year=

  58. [66]

    2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    iCaRL: Incremental Classifier and Representation Learning , author=. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2017

  59. [67]

    ArXiv , year=

    End-to-End Incremental Learning , author=. ArXiv , year=

  60. [68]

    2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Learning a Unified Classifier Incrementally via Rebalancing , author=. 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2019

  61. [69]

    Proceedings of the National Academy of Sciences , year=

    Overcoming catastrophic forgetting in neural networks , author=. Proceedings of the National Academy of Sciences , year=

  62. [70]

    European Conference on Computer Vision , year=

    More Classifiers, Less Forgetting: A Generic Multi-classifier Paradigm for Incremental Learning , author=. European Conference on Computer Vision , year=

  63. [71]

    2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Prototype Augmentation and Self-Supervision for Incremental Learning , author=. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2021

  64. [72]

    Neural Information Processing Systems , year=

    Class-Incremental Learning via Dual Augmentation , author=. Neural Information Processing Systems , year=

  65. [73]

    2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Self-Sustaining Representation Expansion for Non-Exemplar Class-Incremental Learning , author=. 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2022

  66. [74]

    Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

    Fetril: Feature translation for exemplar-free class-incremental learning , author=. Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , pages=

  67. [75]

    International Conference on Machine Learning , year=

    Self-supervised Label Augmentation via Input Transformations , author=. International Conference on Machine Learning , year=

  68. [76]

    Advances in neural information processing systems , volume=

    Supervised contrastive learning , author=. Advances in neural information processing systems , volume=

  69. [77]

    2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Exploring Simple Siamese Representation Learning , author=. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2021

  70. [78]

    CoRR , year=

    Adam: A Method for Stochastic Optimization , author=. CoRR , year=

  71. [79]

    The Eleventh International Conference on Learning Representations , year=

    A Model or 603 Exemplars: Towards Memory-Efficient Class-Incremental Learning , author=. The Eleventh International Conference on Learning Representations , year=

  72. [80]

    Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXIX 16 , pages=

    Label-similarity curriculum learning , author=. Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXIX 16 , pages=. 2020 , organization=

    2020

  73. [81]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Curriculumnet: Weakly supervised learning from large-scale web images , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  74. [82]

    Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XIV 16 , pages=

    Curriculum manager for source selection in multi-source domain adaptation , author=. Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XIV 16 , pages=. 2020 , organization=

    2020

  75. [83]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Der: Dynamically expandable representation for class incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  76. [84]

    Journal of Machine Learning Research , year=

    Visualizing Data using t-SNE , author=. Journal of Machine Learning Research , year=

  77. [85]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

    Learning without Forgetting , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

  78. [86]

    2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Dreaming to Distill: Data-Free Knowledge Transfer via DeepInversion , author=. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2020

  79. [87]

    2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , year=

    Data-Free Network Quantization With Adversarial Knowledge Distillation , author=. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , year=

    2020

  80. [88]

    ArXiv , year=

    Mosaicking to Distill: Knowledge Distillation from Out-of-Domain Data , author=. ArXiv , year=

  81. [89]

    2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    Zero-shot Adversarial Quantization , author=. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , year=

    2021

  82. [90]

    International Conference on Machine Learning , pages=

    Zero-shot knowledge distillation in deep networks , author=. International Conference on Machine Learning , pages=. 2019 , organization=

    2019

  83. [91]

    International Conference on Machine Learning , year=

    Dream distillation: A data-independent model compression framework , author=. International Conference on Machine Learning , year=

  84. [92]

    Neural Information Processing Systems , year=

    Deep Leakage from Gradients , author=. Neural Information Processing Systems , year=

  85. [93]

    Advances in Neural Information Processing Systems , year=

    Data-free knowledge distillation for deep neural networks , author=. Advances in Neural Information Processing Systems , year=

  86. [94]

    ArXiv , year=

    Improved Baselines with Momentum Contrastive Learning , author=. ArXiv , year=

  87. [95]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  88. [96]

    International conference on machine learning , pages=

    Batch normalization: Accelerating deep network training by reducing internal covariate shift , author=. International conference on machine learning , pages=. 2015 , organization=

    2015

  89. [97]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    On the efficacy of knowledge distillation , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  90. [98]

    International Conference on Machine Learning , pages=

    Born again neural networks , author=. International Conference on Machine Learning , pages=. 2018 , organization=

    2018

  91. [99]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Snapshot distillation: Teacher-student optimization in one generation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  92. [100]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Improved knowledge distillation via teacher assistant , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  93. [101]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    A comprehensive overhaul of feature distillation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  94. [102]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Knowledge transfer via distillation of activation boundaries formed by hidden neurons , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  95. [103]

    Advances in neural information processing systems , volume=

    Paraphrasing complex network: Network compression via factor transfer , author=. Advances in neural information processing systems , volume=

  96. [104]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Correlation congruence for knowledge distillation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  97. [105]

    Neurocomputing , volume=

    ISM-Net: Mining incremental semantics for class incremental learning , author=. Neurocomputing , volume=. 2023 , publisher=

    2023

  98. [106]

    Neurocomputing , pages=

    GFR: Generic Feature Representations for Class Incremental Learning , author=. Neurocomputing , pages=. 2023 , publisher=

    2023

  99. [107]

    arXiv preprint arXiv:1412.6572 , year=

    Explaining and harnessing adversarial examples , author=. arXiv preprint arXiv:1412.6572 , year=

    arXiv

  100. [108]

    Advances in neural information processing systems , volume=

    Matching networks for one shot learning , author=. Advances in neural information processing systems , volume=

  101. [109]

    European Conference on Computer Vision , pages=

    Towards recognizing unseen categories in unseen domains , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  102. [110]

    The journal of machine learning research , volume=

    Domain-adversarial training of neural networks , author=. The journal of machine learning research , volume=. 2016 , publisher=

    2016

  103. [111]

    European Conference on Computer Vision , pages=

    R-dfcil: Relation-guided representation learning for data-free class incremental learning , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  104. [112]

    2022 International Joint Conference on Neural Networks (IJCNN) , pages=

    Self-distilled Knowledge Delegator for Exemplar-free Class Incremental Learning , author=. 2022 International Joint Conference on Neural Networks (IJCNN) , pages=. 2022 , organization=

    2022

  105. [113]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Relational knowledge distillation , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  106. [114]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Knowledge distillation via instance relationship graph , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  107. [115]

    Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXVIII 16 , pages=

    Improving knowledge distillation via category structure , author=. Computer Vision--ECCV 2020: 16th European Conference, Glasgow, UK, August 23--28, 2020, Proceedings, Part XXVIII 16 , pages=. 2020 , organization=

    2020

  108. [116]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Exploring inter-channel correlation for diversity-preserved knowledge distillation , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  109. [117]

    International conference on machine learning , pages=

    Continual learning through synaptic intelligence , author=. International conference on machine learning , pages=. 2017 , organization=

    2017

  110. [118]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Self-Organizing Pathway Expansion for Non-Exemplar Class-Incremental Learning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  111. [119]

    The Twelfth International Conference on Learning Representations , year=

    Meta Continual Learning Revisited: Implicitly Enhancing Online Hessian Approximation via Variance Reduction , author=. The Twelfth International Conference on Learning Representations , year=

  112. [120]

    International Conference on Machine Learning , pages=

    Optimizing mode connectivity for class incremental learning , author=. International Conference on Machine Learning , pages=. 2023 , organization=

    2023

  113. [121]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Rehearsal revealed: The limits and merits of revisiting samples in continual learning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  114. [122]

    Advances in neural information processing systems , volume=

    Neural learning in structured parameter spaces-natural Riemannian gradient , author=. Advances in neural information processing systems , volume=

  115. [123]

    Philosophical transactions of the Royal Society of London

    On the mathematical foundations of theoretical statistics , author=. Philosophical transactions of the Royal Society of London. Series A, containing papers of a mathematical or physical character , volume=. 1922 , publisher=

    1922

  116. [124]

    arXiv preprint arXiv:2310.12808 , year=

    Model merging by uncertainty-based gradient matching , author=. arXiv preprint arXiv:2310.12808 , year=

    arXiv

  117. [125]

    International conference on machine learning , pages=

    Essentially no barriers in neural network energy landscape , author=. International conference on machine learning , pages=. 2018 , organization=

    2018

  118. [126]

    Advances in neural information processing systems , volume=

    Loss surfaces, mode connectivity, and fast ensembling of dnns , author=. Advances in neural information processing systems , volume=

  119. [127]

    International Conference on Machine Learning , pages=

    Linear mode connectivity and the lottery ticket hypothesis , author=. International Conference on Machine Learning , pages=. 2020 , organization=

    2020

  120. [128]

    Advances in neural information processing systems , volume=

    What is being transferred in transfer learning? , author=. Advances in neural information processing systems , volume=

  121. [129]

    International Conference on Machine Learning , pages=

    Loss surface simplexes for mode connecting volumes and fast ensembling , author=. International Conference on Machine Learning , pages=. 2021 , organization=

    2021

  122. [130]

    Advances in Neural Information Processing Systems , volume=

    Large scale structure of neural network loss landscapes , author=. Advances in Neural Information Processing Systems , volume=

  123. [131]

    European conference on computer vision , pages=

    Foster: Feature boosting and compression for class-incremental learning , author=. European conference on computer vision , pages=. 2022 , organization=

    2022

  124. [132]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Class-incremental exemplar compression for class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  125. [133]

    European Conference on Computer Vision , pages=

    Dualprompt: Complementary prompting for rehearsal-free continual learning , author=. European Conference on Computer Vision , pages=. 2022 , organization=

    2022

  126. [134]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Learning to prompt for continual learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  127. [135]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Dual-teacher class-incremental learning with data-free generative replay , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  128. [136]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Dual-Consistency Model Inversion for Non-Exemplar Class Incremental Learning , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  129. [137]

    Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

    Adaptive aggregation networks for class-incremental learning , author=. Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition , pages=

  130. [138]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Prototype augmentation and self-supervision for incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  131. [139]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Self-sustaining representation expansion for non-exemplar class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  132. [140]

    9th International Conference on Learning Representations , year=

    Linear Mode Connectivity in Multitask and Continual Learning , author=. 9th International Conference on Learning Representations , year=

  133. [141]

    Psychology of learning and motivation , volume=

    Catastrophic interference in connectionist networks: The sequential learning problem , author=. Psychology of learning and motivation , volume=. 1989 , publisher=

    1989

  134. [142]

    Frontiers in psychology , volume=

    The stability-plasticity dilemma: Investigating the continuum from catastrophic forgetting to age-limited learning effects , author=. Frontiers in psychology , volume=. 2013 , publisher=

    2013

  135. [143]

    Proceedings of the National Academy of Sciences , volume=

    Note on the quadratic penalties in elastic weight consolidation , author=. Proceedings of the National Academy of Sciences , volume=. 2018 , publisher=

    2018

  136. [144]

    Journal of Machine Learning Research , volume=

    New insights and perspectives on the natural gradient method , author=. Journal of Machine Learning Research , volume=

  137. [145]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    FCS: Feature Calibration and Separation for Non-Exemplar Class Incremental Learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  138. [146]

    The Eleventh International Conference on Learning Representations , year=

    Editing models with task arithmetic , author=. The Eleventh International Conference on Learning Representations , year=

  139. [147]

    Advances in neural information processing systems , volume=

    Overcoming catastrophic forgetting by incremental moment matching , author=. Advances in neural information processing systems , volume=

  140. [148]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Online hyperparameter optimization for class-incremental learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  141. [149]

    Advances in Neural Information Processing Systems , volume=

    ACIL: Analytic class-incremental learning with absolute memorization and privacy protection , author=. Advances in Neural Information Processing Systems , volume=

  142. [150]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Gkeal: Gaussian kernel embedded analytic learning for few-shot class incremental task , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  143. [151]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    DS-AL: A dual-stream analytic learning for exemplar-free class-incremental learning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  144. [152]

    Forty-first International Conference on Machine Learning , year=

    Multi-layer rehearsal feature augmentation for class-incremental learning , author=. Forty-first International Conference on Machine Learning , year=

  145. [153]

    Advances in Neural Information Processing Systems , volume=

    Rmm: Reinforced memory management for class-incremental learning , author=. Advances in Neural Information Processing Systems , volume=

  146. [154]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Gradient reweighting: Towards imbalanced class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  147. [155]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Class Incremental Learning with Multi-Teacher Distillation , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  148. [156]

    The Twelfth International Conference on Learning Representations , year=

    Elastic Feature Consolidation For Cold Start Exemplar-Free Incremental Learning , author=. The Twelfth International Conference on Learning Representations , year=

  149. [157]

    Advances in Neural Information Processing Systems , volume=

    S-prompts learning with pre-trained transformers: An occam’s razor for domain incremental learning , author=. Advances in Neural Information Processing Systems , volume=

  150. [158]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Coda-prompt: Continual decomposed attention-based prompting for rehearsal-free continual learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  151. [159]

    The Twelfth International Conference on Learning Representations , year=

    Divide and not forget: Ensemble of selectively trained experts in Continual Learning , author=. The Twelfth International Conference on Learning Representations , year=

  152. [160]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Expandable subspace ensemble for pre-trained model-based class-incremental learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  153. [161]

    European Conference on Computer Vision (ECCV) , year =

    Weighted Ensemble Models Are Strong Continual Learners , author=. European Conference on Computer Vision (ECCV) , year =

  154. [162]

    European Conference on Computer Vision (ECCV) , year =

    MagMax: Leveraging Model Merging for Seamless Continual Learning , author =. European Conference on Computer Vision (ECCV) , year =

  155. [163]

    Forty-second International Conference on Machine Learning , year=

    No Task Left Behind: Isotropic Model Merging with Common and Task-Specific Subspaces , author=. Forty-second International Conference on Machine Learning , year=

  156. [164]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Task singular vectors: Reducing task interference in model merging , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  157. [165]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages=

    Incremental Learning via Robust Parameter Posterior Fusion , author=. Proceedings of the 32nd ACM International Conference on Multimedia , pages=

  158. [166]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Robust fine-tuning of zero-shot models , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

  159. [167]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Rethinking the inception architecture for computer vision , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  160. [168]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Slca: Slow learner with classifier alignment for continual learning on a pre-trained model , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  161. [169]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Finetune like you pretrain: Improved finetuning of zero-shot vision models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  162. [170]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Overcoming generic knowledge loss with selective parameter update , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  163. [171]

    International conference on machine learning , pages=

    Learning transferable visual models from natural language supervision , author=. International conference on machine learning , pages=. 2021 , organization=

    2021

  164. [172]

    Proceedings of the European conference on computer vision (ECCV) , pages=

    Memory aware synapses: Learning what (not) to forget , author=. Proceedings of the European conference on computer vision (ECCV) , pages=

  165. [173]

    Advances in neural information processing systems , volume=

    Continual learning with node-importance based adaptive group sparse regularization , author=. Advances in neural information processing systems , volume=

  166. [174]

    Forty-first International Conference on Machine Learning , year=

    Representation Surgery for Multi-Task Model Merging , author=. Forty-first International Conference on Machine Learning , year=

  167. [175]

    Artificial intelligence and statistics , pages=

    Communication-efficient learning of deep networks from decentralized data , author=. Artificial intelligence and statistics , pages=. 2017 , organization=

    2017

  168. [176]

    Advances in Neural Information Processing Systems , volume=

    Personalized Federated Learning via Feature Distribution Adaptation , author=. Advances in Neural Information Processing Systems , volume=

  169. [177]

    EBioMedicine , volume=

    Decentralised, collaborative, and privacy-preserving machine learning for multi-hospital data , author=. EBioMedicine , volume=. 2024 , publisher=

    2024

  170. [178]

    IEEE Transactions on Big Data , year=

    Privacy-preserving aggregation in federated learning: A survey , author=. IEEE Transactions on Big Data , year=

  171. [179]

    Advances in Neural Information Processing Systems , volume=

    Hugginggpt: Solving ai tasks with chatgpt and its friends in hugging face , author=. Advances in Neural Information Processing Systems , volume=

  172. [180]

    Advances in Neural Information Processing Systems , volume=

    Bridge the modality and capability gaps in vision-language model selection , author=. Advances in Neural Information Processing Systems , volume=

  173. [181]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    Tangent model composition for ensembling and continual fine-tuning , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  174. [182]

    The Thirteenth International Conference on Learning Representations , year=

    A Second-Order Perspective on Model Compositionality and Incremental Learning , author=. The Thirteenth International Conference on Learning Representations , year=

  175. [183]

    Neural computation , volume=

    Adaptive mixtures of local experts , author=. Neural computation , volume=. 1991 , publisher=

    1991

  176. [184]

    Neural computation , volume=

    Hierarchical mixtures of experts and the EM algorithm , author=. Neural computation , volume=. 1994 , publisher=

    1994

  177. [185]

    Task Arithmetic with Lo

    Rajas Chitale and Ankit Vaidya and Aditya Kane and Archana Santosh Ghotkar , booktitle=. Task Arithmetic with Lo

  178. [186]

    Increment Vector Transformation for Class Incremental Learning , author=

  179. [187]

    arXiv preprint arXiv:2412.18219 , year=

    Adapter Merging with Centroid Prototype Mapping for Scalable Class-Incremental Learning , author=. arXiv preprint arXiv:2412.18219 , year=

    arXiv

  180. [188]

    arXiv preprint arXiv:2503.07137 , year=

    A comprehensive survey of mixture-of-experts: Algorithms, theory, and applications , author=. arXiv preprint arXiv:2503.07137 , year=

  181. [189]

    Mixture of Lo

    Xun Wu and Shaohan Huang and Furu Wei , booktitle=. Mixture of Lo

  182. [190]

    arxiv , langid =:2406.03280 , publisher =

    Tang, Anke and Shen, Li and Luo, Yong and Hu, Han and Du, Bo and Tao, Dacheng , year =. arxiv , langid =:2406.03280 , publisher =

  183. [191]

    Advances in Neural Information Processing Systems , volume=

    Beyond not-forgetting: Continual learning with backward knowledge transfer , author=. Advances in Neural Information Processing Systems , volume=

  184. [192]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Preventing zero-shot transfer degradation in continual learning of vision-language models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  185. [193]

    34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 , pages=

    Averaging weights leads to wider optima and better generalization , author=. 34th Conference on Uncertainty in Artificial Intelligence 2018, UAI 2018 , pages=. 2018 , organization=

    2018

  186. [194]

    European Conference on Computer Vision , pages=

    Adapt without forgetting: Distill proximity from dual teachers in vision-language models , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  187. [195]

    European Conference on Computer Vision , pages=

    Mind the interference: Retaining pre-trained knowledge in parameter efficient continual learning of vision-language models , author=. European Conference on Computer Vision , pages=. 2024 , organization=

    2024

  188. [196]

    2010 IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Sun database: Large-scale scene recognition from abbey to zoo , author=. 2010 IEEE Conference on Computer Vision and Pattern Recognition , pages=. 2010 , organization=

    2010

  189. [197]

    2013 IEEE International Conference on Computer Vision Workshops , pages=

    3D object representations for fine-grained categorization , author=. 2013 IEEE International Conference on Computer Vision Workshops , pages=. 2013 , organization=

    2013

  190. [198]

    Proceedings of the IEEE , volume=

    Remote sensing image scene classification: Benchmark and state of the art , author=. Proceedings of the IEEE , volume=. 2017 , publisher=

    2017

  191. [199]

    IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification , author=. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , volume=. 2019 , publisher=

    2019

  192. [200]

    NIPS Workshop on Deep Learning and Unsupervised Feature Learning , year=

    Reading digits in natural images with unsupervised feature learning , author=. NIPS Workshop on Deep Learning and Unsupervised Feature Learning , year=

  193. [201]

    computer: Benchmarking machine learning algorithms for traffic sign recognition , author=

    Man vs. computer: Benchmarking machine learning algorithms for traffic sign recognition , author=. Neural Networks , volume=. 2012 , publisher=

    2012

  194. [202]

    Proceedings of the IEEE , volume=

    Gradient-based learning applied to document recognition , author=. Proceedings of the IEEE , volume=. 1998 , publisher=

    1998

  195. [203]

    Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Describing textures in the wild , author=. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition , pages=

  196. [204]

    2008 Sixth Indian Conference on Computer Vision, Graphics & Image Processing , pages=

    Automated flower classification over a large number of classes , author=. 2008 Sixth Indian Conference on Computer Vision, Graphics & Image Processing , pages=. 2008 , organization=

    2008

  197. [205]

    arXiv preprint arXiv:1806.03962 , year=

    Rotation equivariant CNNs for digital pathology , author=. arXiv preprint arXiv:1806.03962 , year=

    arXiv

  198. [206]

    International Conference on Neural Information Processing , pages=

    Challenges in representation learning: A report on three machine learning contests , author=. International Conference on Neural Information Processing , pages=. 2013 , organization=

    2013

  199. [207]

    2012 IEEE Conference on Computer Vision and Pattern Recognition , pages=

    Cats and dogs , author=. 2012 IEEE Conference on Computer Vision and Pattern Recognition , pages=. 2012 , publisher=

    2012

  200. [208]

    Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics , pages=

    An analysis of single-layer networks in unsupervised feature learning , author=. Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics , pages=. 2011 , organization=

    2011

  201. [209]

    Learning multiple layers of features from tiny images , author=

  202. [210]

    European Conference on Computer Vision , pages=

    Food-101--mining discriminative components with random forests , author=. European Conference on Computer Vision , pages=. 2014 , organization=

    2014

  203. [211]

    arXiv preprint arXiv:1708.07747 , year=

    Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms , author=. arXiv preprint arXiv:1708.07747 , year=

    arXiv

  204. [212]

    2017 International Joint Conference on Neural Networks (IJCNN) , pages=

    EMNIST: Extending MNIST to handwritten letters , author=. 2017 International Joint Conference on Neural Networks (IJCNN) , pages=. 2017 , organization=

    2017

  205. [213]

    NeurIPS Workshop on Machine Learning for Creativity and Design , year=

    Deep learning for classical Japanese literature , author=. NeurIPS Workshop on Machine Learning for Creativity and Design , year=

  206. [214]

    Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing , pages=

    Recursive deep models for semantic compositionality over a sentiment treebank , author=. Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing , pages=

    2013

  207. [215]

    Journal of Machine Learning Research , volume=

    Scaling instruction-finetuned language models , author=. Journal of Machine Learning Research , volume=

  208. [216]

    7th International Conference on Learning Representations, ICLR 2019 , year=

    Glue: A multi-task benchmark and analysis platform for natural language understanding , author=. 7th International Conference on Learning Representations, ICLR 2019 , year=

    2019

  209. [217]

    Forty-second International Conference on Machine Learning , year=

    Whoever Started the interference Should End It: Guiding Data-Free Model Merging via Task Vectors , author=. Forty-second International Conference on Machine Learning , year=

  210. [218]

    Forty-first International Conference on Machine Learning , year=

    Localizing Task Information for Improved Model Merging and Compression , author=. Forty-first International Conference on Machine Learning , year=

  211. [219]

    Academic Press , year=

    Matrix perturbation theory , author=. Academic Press , year=

  212. [220]

    BIT Numerical Mathematics , volume=

    Perturbation bounds in connection with singular value decomposition , author=. BIT Numerical Mathematics , volume=. 1972 , publisher=

    1972

  213. [221]

    Mathematische Annalen , volume=

    Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differentialgleichungen (mit einer Anwendung auf die Theorie der Hohlraumstrahlung) , author=. Mathematische Annalen , volume=. 1912 , publisher=

    1912

  214. [222]

    Zihuan Qiu and Yi Xu and Chiyuan He and Fanman Meng and Linfeng Xu and Qingbo Wu and Hongliang Li , booktitle=

  215. [223]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    MetaGPT: Merging Large Language Models Using Model Exclusive Task Arithmetic , author=. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    2024

  216. [224]

    Advances in neural information processing systems , volume=

    Visual instruction tuning , author=. Advances in neural information processing systems , volume=

  217. [225]

    Advances in Neural Information Processing Systems , volume=

    Learn to explain: Multimodal reasoning via thought chains for science question answering , author=. Advances in Neural Information Processing Systems , volume=

  218. [226]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Vizwiz grand challenge: Answering visual questions from blind people , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  219. [227]

    arXiv preprint arXiv:2110.13214 , year=

    Iconqa: A new benchmark for abstract diagram understanding and visual language reasoning , author=. arXiv preprint arXiv:2110.13214 , year=

    arXiv

  220. [228]

    2009 IEEE conference on computer vision and pattern recognition , pages=

    Imagenet: A large-scale hierarchical image database , author=. 2009 IEEE conference on computer vision and pattern recognition , pages=. 2009 , organization=

    2009

  221. [229]

    The Thirteenth International Conference on Learning Representations , year=

    AlphaEdit: Null-Space Constrained Model Editing for Language Models , author=. The Thirteenth International Conference on Learning Representations , year=

  222. [230]

    arXiv preprint arXiv:2411.18729 , year=

    Multi-task model merging via adaptive weight disentanglement , author=. arXiv preprint arXiv:2411.18729 , year=

    arXiv

  223. [231]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    Orthogonal subspace learning for language model continual learning , author=. Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    2023

  224. [232]

    International conference on artificial intelligence and statistics , pages=

    Orthogonal gradient descent for continual learning , author=. International conference on artificial intelligence and statistics , pages=. 2020 , organization=

    2020

  225. [233]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Inflora: Interference-free low-rank adaptation for continual learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  226. [234]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Wang, Shipeng and Li, Xiaorong and Sun, Jian and Xu, Zongben , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2021 , pages =

    2021

  227. [235]

    Modeling Multi-Task Model Merging as Adaptive Projective Gradient Descent , author=

  228. [236]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

    The Many Faces of Robustness: A Critical Analysis of Out-of-Distribution Generalization , author=. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages=

  229. [237]

    Advances in Neural Information Processing Systems , volume=

    ObjectNet: A Large-Scale Bias-Controlled Dataset for Pushing the Limits of Object Recognition Models , author=. Advances in Neural Information Processing Systems , volume=

  230. [238]

    The Fourteenth International Conference on Learning Representations , year=

    Null-Space Filtering for Data-free Continual Model Merging: Preserving Stability, Promoting Plasticity , author=. The Fourteenth International Conference on Learning Representations , year=

  231. [239]

    International Conference on Machine Learning , pages=

    Model Tailor: Mitigating Catastrophic Forgetting in Multi-modal Large Language Models , author=. International Conference on Machine Learning , pages=. 2024 , organization=

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.