{"total":81,"items":[{"citing_arxiv_id":"2606.17545","ref_index":28,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Continuous-time Optimal Stopping through Deep Reinforcement Learning","primary_cat":"cs.LG","submitted_at":"2026-06-16T05:49:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"CARLOS employs an aggregate deep neural network trained on progressively finer time grids with adaptive sampling to learn continuous-time exercise boundaries for optimal stopping, delivering higher values than discrete Bermudan methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.00400","ref_index":38,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Dynamic Proxy-Mixing: Transferring Replay Controllers from Small to Large Models for Continual Instruction Tuning","primary_cat":"cs.LG","submitted_at":"2026-05-29T22:32:55+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"PROXYMIX learns a dynamic replay controller on a small proxy model and transfers it to a large target model, improving accuracy by 3.4 points and reducing forgetting by 3.5 points on LLaMA-3-8B continual tuning sequences.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.00382","ref_index":26,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"CRMA: A Spectrally-Bounded Backbone for Modular Continual Fine-Tuning of LLMs","primary_cat":"cs.LG","submitted_at":"2026-05-29T21:50:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"CRMA adds a spectrally bounded residual adapter backbone to modular continual fine-tuning of LLMs, achieving near-zero loss drift and positive backward transfer on Mistral-7B across domains.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.30537","ref_index":8,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"The Long-Term Effects of Data Selection in LLM Fine-Tuning","primary_cat":"cs.LG","submitted_at":"2026-05-28T20:12:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Short-term data selectors in multi-stage LLM fine-tuning can slow future learning and increase forgetting, formalized as myopic selection with a proposed LHAS objective to address it.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.29380","ref_index":17,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"TRACER: Persistent Regularization for Robust Multimodal Finetuning","primary_cat":"cs.LG","submitted_at":"2026-05-28T05:34:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"TRACER applies weighted moving average distillation in contrastive finetuning of multimodal models to retain pretrained knowledge and boost out-of-distribution accuracy.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.28495","ref_index":22,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Janus-LoRA: A Balanced Low-Rank Adaptation for Continual Learning","primary_cat":"cs.CV","submitted_at":"2026-05-27T13:53:14+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Janus-LoRA uses gradient rectification via online subspace estimation and a decoupled margin loss to enforce parameter orthogonality and feature separation in LoRA-based continual learning, reporting new SOTA results.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.27762","ref_index":4,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"PEAM: Parametric Embodied Agent Memory through Contrastive Internalization of Experience in Minecraft","primary_cat":"cs.AI","submitted_at":"2026-05-26T23:20:58+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"PEAM is a parametric memory framework for Minecraft agents that internalizes experiences into a multimodal MoE-LoRA module using contrastive objectives on failures and a scale-free self-triggered consolidation mechanism.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.23565","ref_index":58,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Understanding Goal Generalisation in Sequential Reinforcement Learning","primary_cat":"cs.LG","submitted_at":"2026-05-22T12:31:18+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Empirical analysis of over 100 sequential RL training pipelines across 250+ OOD environments finds salient features drive generalization and early goals persist, with latent policy gradients simulating latent variable evolution to predict OOD behavior from training history.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.22035","ref_index":26,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"HyLoVQA: Dynamic Hypernetwork-Generated Low-Rank Adaptation for Continual Visual Question Answering","primary_cat":"cs.CV","submitted_at":"2026-05-21T06:12:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"HyLoVQA combines an anchor memory bank with hypernetwork-generated LoRA adapters and an alignment loss to adapt to new VQA tasks while limiting interference with prior knowledge.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20803","ref_index":18,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Tunable MAGMAX: Preference-Aware Model Merging for Continual Learning","primary_cat":"cs.LG","submitted_at":"2026-05-20T06:50:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Tunable MAGMAX adds a tunable preference vector to model merging for continual learning, enabling automatic adaptation to target environments using small amounts of data while maintaining or improving task-wise performance.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20297","ref_index":6,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"MedCRP-CL: Continual Medical Image Segmentation via Bayesian Nonparametric Semantic Modality Discovery","primary_cat":"cs.CV","submitted_at":"2026-05-19T11:28:16+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"MedCRP-CL discovers semantic modalities online via CRP from text prompts and maintains modality-specific LoRA adapters with intra-modality EWC, achieving 73.3% Dice and 4.1% forgetting on 16 tasks while using 6x fewer parameters than the best baseline.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.20247","ref_index":10,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"CP-MoE: Consistency-Preserving Mixture-of-Experts for Continual Learning","primary_cat":"cs.LG","submitted_at":"2026-05-18T06:31:14+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"CP-MoE uses a transient expert, consistency-preserving routing bias, and guided regularization to reduce catastrophic forgetting in MoE-based LLMs and VLMs while preserving cross-task transfer, reporting SOTA on SuperNI and gains on VQA v2.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.16967","ref_index":62,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Expandable, Compressible, Mineable: Open-World Thermal Image Restoration","primary_cat":"cs.CV","submitted_at":"2026-05-16T12:41:38+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"ECMRNet is a continual-learning restoration network that decomposes features into isolated groups, expands new groups for novel degradations, prunes via structural entropy, and mines historical components for compound degradations in open-world TIR imaging.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.16844","ref_index":15,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Artificial Adaptive Intelligence: The Missing Stage Between Narrow and General Intelligence","primary_cat":"cs.AI","submitted_at":"2026-05-16T07:04:28+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Proposes Artificial Adaptive Intelligence as the regime between narrow and general AI, defined by elimination of human-specified hyperparameters, and introduces an adaptivity index plus parametric minimality principle grounded in minimum description length.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.16757","ref_index":25,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning","primary_cat":"cs.AI","submitted_at":"2026-05-16T02:11:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"NeuroMAS reframes multi-agent language systems as neural architectures where LLM agents learn coordination via reinforcement learning rather than predefined roles.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15775","ref_index":89,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Continual Learning of Domain-Invariant Representations","primary_cat":"cs.LG","submitted_at":"2026-05-15T09:31:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Introduces replay-based continual learning with sequential invariance alignment to learn domain-invariant representations, outperforming baselines on generalization to unseen domains across six datasets in vision, medicine, manufacturing, and ecology.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15435","ref_index":37,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"On the Stability of Growth in Structural Plasticity","primary_cat":"cs.LG","submitted_at":"2026-05-14T21:27:38+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.15053","ref_index":17,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"TFGN: Task-Free, Replay-Free Continual Pre-Training Without Catastrophic Forgetting at LLM Scale","primary_cat":"cs.LG","submitted_at":"2026-05-14T16:46:26+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"TFGN is an architectural overlay for transformers enabling task-free, replay-free continual pre-training across heterogeneous domains at LLM scale with near-zero backward transfer and high gradient orthogonality.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.14364","ref_index":24,"ref_count":3,"confidence":0.98,"is_internal_anchor":true,"paper_title":"MoRe: Modular Representations for Principled Continual Representation Learning on Sequential Data","primary_cat":"cs.LG","submitted_at":"2026-05-14T04:46:54+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"MoRe identifies modular structure in representations themselves to enable principled reuse, alignment, and expansion of modules during continual adaptation on sequential data.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.14304","ref_index":3,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Matrix-Space Reinforcement Learning for Reusing Local Transition Geometry","primary_cat":"cs.LG","submitted_at":"2026-05-14T03:12:29+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"MSRL represents trajectory segments as PSD matrices to prove additive composition properties and bootstrap value functions for better transfer, reaching 0.73 AUC versus 0.57-0.65 baselines.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12789","ref_index":13,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Lifelong Learning in Vision-Language Models: Enhanced EWC with Cross-Modal Knowledge Retention","primary_cat":"cs.RO","submitted_at":"2026-05-12T22:05:30+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Enhanced EWC for LVLMs cuts forgetting rates by 78% versus naive training and keeps visual-textual alignment with 15% extra compute.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.12306","ref_index":14,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"KAN-CL: Per-Knot Importance Regularization for Continual Learning with Kolmogorov-Arnold Networks","primary_cat":"cs.LG","submitted_at":"2026-05-12T15:55:09+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":7.0,"formal_verification":"none","one_line_summary":"KAN-CL cuts catastrophic forgetting by 88-93% on Split-CIFAR-10/5T and Split-CIFAR-100/10T by anchoring KAN parameters at per-knot granularity while matching baseline accuracy.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Replay-based CL.iCaRL [8], ER [9], GEM [10], and DER++ [11] use stored or distilled examples to rehearse old tasks. Replay is complementary to regularization; we show thatKAN-CL+replay with anchor annealing composes well in class-IL settings. Parameter isolation and expansion.Progressive Neural Networks add new columns per task and share via lateral connections [14]; PackNet iteratively prunes and reuses weights [15]. These approaches require task identities, mask management, or growing capacity. Our method keeps a fixed shared head and exploits locality already present in the spline basis. Hybrid backbone-head CL.Using a frozen or slowly adapting backbone with a task-specific head is a well- studied strategy [16]."},{"citing_arxiv_id":"2605.11617","ref_index":55,"ref_count":2,"confidence":0.98,"is_internal_anchor":true,"paper_title":"MIST: Reliable Streaming Decision Trees for Online Class-Incremental Learning via McDiarmid Bound","primary_cat":"cs.LG","submitted_at":"2026-05-12T06:45:00+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"MIST fixes unreliable splits in streaming decision trees for class-incremental learning by replacing Hoeffding-style bounds with a K-independent McDiarmid radius on Gini, plus Bayesian parent-to-child inheritance and per-leaf quantile sketches.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"[53] Sylvestre-Alvise Rebuffi, Alexander Kolesnikov, Georg Sperl, and Christoph H. Lampert. icarl: Incremental classifier and representation learning. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), July 2017. [54] David Rolnick, Arun Ahuja, Jonathan Schwarz, Timothy P. Lillicrap, and Greg Wayne.Experi- ence replay for continual learning. Curran Associates Inc., Red Hook, NY , USA, 2019. [55] Andrei Rusu, Neil Rabinowitz, Guillaume Desjardins, Hubert Soyer, James Kirkpatrick, Koray Kavukcuoglu, Razvan Pascanu, and Raia Hadsell. Progressive neural networks.arXiv preprint arXiv:1606.04671, 06 2016. [56] Leszek Rutkowski, Lena Pietruczuk, Piotr Duda, and Maciej Jaworski. Decision trees for mining data streams based on the mcdiarmid's bound."},{"citing_arxiv_id":"2605.11369","ref_index":44,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Dynamic Full-body Motion Agent with Object Interaction via Blending Pre-trained Modular Controllers","primary_cat":"cs.CV","submitted_at":"2026-05-12T00:43:26+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"A two-stage framework augments HOI data with dynamic priors and blends pre-trained dynamic motion and static interaction agents via a composer network to enable long-term dynamic human-object interactions with higher success rates and reduced training time.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"ing dynamic HOI tasks such as object carrying in desired trajectories via consistent contacts. Blending with pretrained Expert Models.To tackle com- plex tasks, recent works leverage pretrained expert models that have been trained on similar tasks, rather than learn- ing from scratch. Residual Policy Learning (RPL) [13, 47, 60, 63] and Progressive Neural Networks (PNN) [44] en- able adaptation of an existing model using additional neural networks. In contrast, Mixture of Experts (MoE) [2, 11, 46] allows dynamic selection or combination of multiple expert models for new tasks, effectively leveraging the capabilities of existing experts. While MoE [46] typically uses sparse gating or hard selection, recent approaches [11] explore soft"},{"citing_arxiv_id":"2605.09355","ref_index":46,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"FLAME: Adaptive Mixture-of-Experts for Continual Multimodal Multi-Task Learning","primary_cat":"cs.LG","submitted_at":"2026-05-10T06:09:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"FLAME is an MoE architecture using modality-specific routers and low-rank compression of expert knowledge to support efficient continual multimodal multi-task learning while reducing catastrophic forgetting.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"[44] Nicola Rieke, Jonny Hancox, Wenqi Li, Fausto Milletari, Holger R Roth, Shadi Albarqouni, Spyridon Bakas, Mathieu N Galtier, Bennett A Landman, Klaus Maier-Hein, et al. The future of digital health with federated learning.NPJ digital medicine, 3(1):119, 2020. [45] Sebastian Ruder. An overview of multi-task learning in deep neural networks.arXiv preprint arXiv:1706.05098, 2017. [46] Andrei A Rusu, Neil C Rabinowitz, Guillaume Desjardins, Hubert Soyer, James Kirkpatrick, Koray Kavukcuoglu, Razvan Pascanu, and Raia Hadsell. Progressive neural networks.arXiv preprint arXiv:1606.04671, 2016. [47] Grzegorz Rype's'c, Sebastian Cygert, Valeriya Khan, Tomasz Trzci'nski, Bartosz Zieli'nski, and Bartłomiej Twardowski. Divide and not forget: Ensemble of selectively trained experts in"},{"citing_arxiv_id":"2605.08311","ref_index":56,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Revitalizing the Beginning: Avoiding Storage Dependency for Model Merging in Continual Learning","primary_cat":"cs.LG","submitted_at":"2026-05-08T14:07:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"The paper proposes Trajectory Regularized Merging (TRM) to enable storage-free model merging in continual learning by optimizing in an augmented trajectory subspace with task alignment, prediction consistency, and gradient responsiveness objectives, claiming SOTA results.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.07494","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"DIMoE-Adapters: Dynamic Expert Evolution for Continual Learning in Vision-Language Models","primary_cat":"cs.CV","submitted_at":"2026-05-08T09:32:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"DIMoE-Adapters uses self-calibrated expert evolution and prototype-guided selection to dynamically grow and allocate experts, outperforming prior continual learning methods on vision-language models.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"gation and zero-shot preservation. 2 Related Work Continual Learning.Existing continual learning methods can be broadly categorized into regularization-based, architecture-based, and replay-based approaches. Regularization-based meth- ods [12, 4] constrain parameter updates by importance estimation, but may overly restrict model plasticity. Architecture-based methods [13, 14] allocate task-specific parameters to reduce forgetting, but may hinder knowledge transfer across tasks. Replay-based methods [15, 16] mitigate forgetting by rehearsing stored or generated samples. However, most existing approaches focus on a single incremental setting, such as class-incremental [17] or domain-incremental [18] learning, which limits"},{"citing_arxiv_id":"2605.07038","ref_index":84,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Learning Material-Aware Hamiltonian Risk Fields for Safe Navigation","primary_cat":"cs.LG","submitted_at":"2026-05-07T23:33:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"A learned context-energy term in port-Hamiltonian policies creates selective risk navigation that activates evasive forces only when safer paths are available.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.06160","ref_index":22,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Beyond Forgetting in Continual Medical Image Segmentation: A Comprehensive Benchmark Study","primary_cat":"cs.CV","submitted_at":"2026-05-07T12:48:56+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Benchmark experiments in continual medical image segmentation reveal that no single method satisfies all clinical requirements, with replay-based approaches offering the best stability-plasticity trade-off while forward generalizability needs more attention.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.05495","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Shortcut Solutions Learned by Transformers Impair Continual Compositional Reasoning","primary_cat":"cs.LG","submitted_at":"2026-05-06T22:31:59+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"BERT learns shortcut solutions that impair generalization and forward transfer in continual LEGO, while ALBERT learns loop-like solutions for better performance, yet both fail at cross-experience composition, with ALBERT rescued by mixed-data training.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.03832","ref_index":13,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"A Domain Incremental Continual Learning Benchmark for ICU Time Series Model Transportability","primary_cat":"cs.LG","submitted_at":"2026-05-05T15:02:07+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Proposes a domain incremental continual learning benchmark for ICU time series model transportability across US regions and evaluates data replay and EWC methods.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.03555","ref_index":24,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"MILE: Mixture of Incremental LoRA Experts for Continual Semantic Segmentation across Domains and Modalities","primary_cat":"cs.CV","submitted_at":"2026-05-05T09:27:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"MILE combines incremental LoRA experts with prototype-guided gating to support continual semantic segmentation across domains and modalities while adding only a small number of parameters per task.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.02509","ref_index":8,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"MPCS: Neuroplastic Continual Learning via Multi-Component Plasticity and Topology-Aware EWC","primary_cat":"cs.LG","submitted_at":"2026-05-04T12:04:09+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"MPCS integrates eleven plasticity mechanisms and reaches a Normalized Efficiency Score of 94.2 on a 31-task benchmark, with ablations showing that removing EWC and Hebbian updates yields higher performance at lower cost.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.02105","ref_index":74,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Sharpness-Aware Pretraining Mitigates Catastrophic Forgetting","primary_cat":"cs.LG","submitted_at":"2026-05-04T00:02:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Sharpness-aware pretraining and related flat-minima interventions reduce catastrophic forgetting by up to 80% after post-training across 20M-150M models and by 31-40% at 1B scale.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.27031","ref_index":19,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"NORACL: Neurogenesis for Oracle-free Resource-Adaptive Continual Learning","primary_cat":"cs.LG","submitted_at":"2026-04-29T15:04:25+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"NORACL dynamically grows network capacity via neurogenesis-inspired signals to achieve oracle-level continual learning performance without pre-specifying architecture size.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.24637","ref_index":34,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Cortex-Inspired Continual Learning: Unsupervised Instantiation and Recovery of Functional Task Networks","primary_cat":"cs.LG","submitted_at":"2026-04-27T16:06:28+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"FTN achieves near-zero forgetting on continual learning benchmarks by isolating task subnetworks via self-organizing binary masks generated through gradient descent, smoothing, and k-winner-take-all.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.18857","ref_index":12,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Task Switching Without Forgetting via Proximal Decoupling","primary_cat":"cs.LG","submitted_at":"2026-04-20T21:28:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Operator splitting separates task optimization from proximal stability enforcement to achieve forgetting-free continual learning with SOTA benchmark results.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.15794","ref_index":5,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Self-Distillation as a Performance Recovery Mechanism for LLMs: Counteracting Compression and Catastrophic Forgetting","primary_cat":"cs.LG","submitted_at":"2026-04-17T07:55:38+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Self-distillation fine-tuning recovers LLM capabilities by aligning the student's high-dimensional hidden-layer manifold with the teacher's, as quantified by CKA correlation with performance gains.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.14259","ref_index":34,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Continual Learning for fMRI-Based Brain Disorder Diagnosis via Functional Connectivity Matrices Generative Replay","primary_cat":"q-bio.TO","submitted_at":"2026-04-15T16:08:52+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"A structure-aware VAE generates realistic FC matrices for replay, combined with multi-level knowledge distillation and hierarchical contextual bandit sampling, to enable continual fMRI-based brain disorder diagnosis across sequentially arriving multi-site data without catastrophic forgetting.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.13730","ref_index":33,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"ReConText3D: Replay-based Continual Text-to-3D Generation","primary_cat":"cs.CV","submitted_at":"2026-04-15T11:11:36+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":8.0,"formal_verification":"none","one_line_summary":"ReConText3D is the first replay-memory framework for continual text-to-3D generation that prevents catastrophic forgetting on new textual categories while preserving quality on previously seen classes.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.12167","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"EMBER: Autonomous Cognitive Behaviour from Learned Spiking Neural Network Dynamics in a Hybrid LLM Architecture","primary_cat":"cs.AI","submitted_at":"2026-04-14T00:51:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"A hybrid SNN-LLM system uses learned spiking dynamics and lateral STDP propagation to trigger LLM actions without external prompts, producing the first autonomous action after 7 exchanges from a clean start.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.10947","ref_index":27,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Multi-Faceted Continual Knowledge Graph Embedding for Semantic-Aware Link Prediction","primary_cat":"cs.IR","submitted_at":"2026-04-13T03:41:01+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"MF-CKGE separates temporal old and new knowledge into distinct embedding spaces with semantic decoupling and adaptive importance scoring to improve continual link prediction.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.10549","ref_index":5,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Failure Ontology: A Lifelong Learning Framework for Blind Spot Detection and Resilience Design","primary_cat":"cs.AI","submitted_at":"2026-04-12T09:39:53+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Failure Ontology offers a four-type taxonomy of blind spots, five failure patterns, and a theorem claiming failure-based learning is more sample-efficient than success-based learning under limited data.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.09452","ref_index":31,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"SafeAdapt: Provably Safe Policy Updates in Deep Reinforcement Learning","primary_cat":"cs.LG","submitted_at":"2026-04-10T16:09:39+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SafeAdapt certifies a Rashomon set of safe policies from demonstration data and projects updates from arbitrary RL algorithms onto it to guarantee preservation of safety on source tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.08159","ref_index":28,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Face-D(^2)CL: Multi-Domain Synergistic Representation with Dual Continual Learning for Facial DeepFake Detection","primary_cat":"cs.CV","submitted_at":"2026-04-09T12:18:42+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"Face-D²CL fuses spatial and frequency features and uses dual continual learning to reduce forgetting while adapting to new DeepFakes, cutting average error rates by 60.7% and raising unseen-domain AUC by 7.9% over prior SOTA.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.07799","ref_index":29,"ref_count":2,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Learning Without Losing Identity: Capability Evolution for Embodied Agents","primary_cat":"cs.RO","submitted_at":"2026-04-09T04:51:07+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Embodied agents maintain persistent identity while evolving modular capabilities through a closed-loop process, raising simulated task success from 32.4% to 91.3% with zero policy drift.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.07108","ref_index":6,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Information as Structural Alignment: A Dynamical Theory of Continual Learning","primary_cat":"cs.LG","submitted_at":"2026-04-08T14:00:14+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"IBF achieves near-zero forgetting and positive backward transfer in continual learning by driving configurations toward coherence through motion and modification dynamics without storing raw data.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.06425","ref_index":29,"ref_count":1,"confidence":0.9,"is_internal_anchor":true,"paper_title":"Neural Computers","primary_cat":"cs.LG","submitted_at":"2026-04-07T20:01:05+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Neural Computers are introduced as a new machine form where computation, memory, and I/O are unified in a learned runtime state, with initial video-model experiments showing acquisition of basic interface primitives from traces.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.02778","ref_index":39,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"When Modalities Remember: Continual Learning for Multimodal Knowledge Graphs","primary_cat":"cs.CL","submitted_at":"2026-04-03T06:40:04+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"MRCKG combines a multimodal-structural curriculum, cross-modal preservation, and contrastive replay to let multimodal knowledge graphs learn new entities and relations over time without catastrophic forgetting.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"veal an important finding: multimodal information can serve as a semantic anchor in continual knowledge graph reasoning. 2 Related Work 2.1 Multimodal Knowledge Graph Reasoning MMKGR explicitly incorporates visual[45] and textual information into entity representation learning to make up for the limits of purely structural methods[7]. In multimodal fusion, MKGformer[6], IMF[25], and LAFA[39] explore fusion strategies from the perspec- tives of cross-modal Transformers, interaction mechanisms, and neighbor structure information, respectively. More recent studies further improve fusion granularity and robustness, including the fine-grained tokenization of MYGO[57], the attention penalty of APKGC[11], the frequency-domain fusion of WFF[52], the dynamic"},{"citing_arxiv_id":"2604.13085","ref_index":6,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Adaptive Memory Crystallization for Autonomous AI Agent Learning in Dynamic Environments","primary_cat":"cs.LG","submitted_at":"2026-04-02T22:53:34+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"AMC models memory consolidation via a Liquid-Glass-Crystal process governed by an SDE with proven convergence to a Beta distribution, yielding 34-43% better forward transfer and 67-80% less forgetting on standard continual RL benchmarks.","context_count":1,"top_context_role":"baseline","top_context_polarity":"baseline","context_text":"on a fixed-size experience-replay buffer [2] combined with stochastic gradient descent. When the task distribution shifts, gradient updates on new data overwrite weights encoding old behaviors-a phenomenon known ascatastrophic forgetting [3]. Existing mitigations fall into three families:regulariza- tionmethods [4], [5], which add parameter-protection penal- ties;dynamic architecturemethods [6], [7], which grow or partition network capacity; andmemory-replaymethods [8]- [10], which curate which experiences to replay. Each family has known failure modes for long-lived agents (Section II). Neuroscience offers a complementary design principle: synaptic consolidation[11]. Under the Synaptic Tagging and Capture (STC) hypothesis [12], early-phase LTP pro-"}],"limit":50,"offset":0}