pith. sign in

arxiv: 2605.14948 · v1 · pith:BBGX6M5Rnew · submitted 2026-05-14 · 💻 cs.CV

ACE-LoRA: Adaptive Orthogonal Decoupling for Continual Image Editing

Yuehao Liu , Weijia Zhang , Xuanming Shang , Zhizhou Chen , Yanhao Ge , Shanyan Guan , Chao Ma This is my paper

Pith reviewed 2026-06-30 21:43 UTC · model grok-4.3

classification 💻 cs.CV
keywords continual learningimage editingdiffusion modelsLoRAcatastrophic forgettingorthogonal decouplingCIE-Benchparameter-efficient fine-tuning
0
0 comments X

The pith

Adaptive orthogonal decoupling lets diffusion models learn new image edits without forgetting old ones.

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

The paper introduces ACE-LoRA as a method to handle continual learning in image editing tasks using diffusion models. It identifies interfering updates between tasks and makes them orthogonal to reduce conflicts in the parameter space. A compression technique preserves information from past tasks in a way that does not change the rank of the adaptations. The authors also create CIE-Bench to test these methods across various editing scenarios. If successful, this would allow models to be updated over time for new editing instructions while keeping their ability to handle earlier ones.

Core claim

ACE-LoRA mitigates catastrophic forgetting in continual image editing by using Adaptive Orthogonal Decoupling to identify and orthogonalize task interference and Rank-Invariant Historical Information Compression to maintain scalability, leading to improved instruction fidelity, visual realism, and robustness compared to existing approaches on the CIE-Bench benchmark.

What carries the argument

Adaptive Orthogonal Decoupling, which detects task interference and enforces orthogonality between task-specific parameter updates to minimize forgetting.

Load-bearing premise

That interfering directions between different editing tasks can be accurately identified in the low-rank parameter space and made orthogonal without reducing the effectiveness of the adaptations for any task.

What would settle it

Training the model on a series of sequential editing tasks and then measuring a substantial decline in performance on the first tasks relative to a model trained only on those first tasks would falsify the effectiveness of the decoupling.

Figures

Figures reproduced from arXiv: 2605.14948 by Chao Ma, Shanyan Guan, Weijia Zhang, Xuanming Shang, Yanhao Ge, Yuehao Liu, Zhizhou Chen.

Figure 1
Figure 1. Figure 1: (a)&(b) Analysis on LoRA similarities between tasks under individual/sequential finetuning. (c)&(d) Analysis on SVD energy proportion/reconstruction error for history compression. Existing works for continual learning can be broadly categorized into architecture-based, rehearsal￾based, and regularization-based methods. Architecture-based methods [3, 13, 19, 49] expand the model with task-specific modules t… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of ACE-LoRA. ACE-LoRA leverages Adaptive Orthogonal Decoupling to [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of CIE-Bench for continual image editing. CIE-Bench consists of three main [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison between our evaluation metrics and ImgEdit-Judge [52]. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative comparison of visual results for different methods on CIE-Bench. For each [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Visualization of CIIE-Bench, which consists of six sub-tasks: ERP Outpainting, Refocus, [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
read the original abstract

State-of-the-art diffusion models often rely on parameter-efficient fine-tuning to perform specialized image editing tasks. However, real-world applications require continual adaptation to new tasks while preserving previously learned knowledge. Despite the practical necessity, continual learning for image editing remains largely underexplored. We propose ACE-LoRA, a dynamic regularization framework for continual image editing that effectively mitigates catastrophic forgetting. ACE-LoRA leverages Adaptive Orthogonal Decoupling to identify and orthogonalize task interference, and introduces a Rank-Invariant Historical Information Compression strategy to address scalability issues in continual updates. To facilitate continual learning in image editing and provide a standardized evaluation protocol, we introduce CIE-Bench, the first comprehensive benchmark in this domain. CIE-Bench encompasses diverse and practically relevant image editing scenarios with a balanced level of difficulty to effectively expose limitations of existing models while remaining compatible with parameter-efficient fine-tuning. Extensive experiments demonstrate that our method consistently outperforms existing baselines in terms of instruction fidelity, visual realism, and robustness to forgetting, establishing a strong foundation for continual learning in image editing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes ACE-LoRA, a dynamic regularization framework for continual image editing with diffusion models that uses Adaptive Orthogonal Decoupling to identify and orthogonalize task interference in LoRA parameter space together with Rank-Invariant Historical Information Compression for scalability; it also introduces the CIE-Bench benchmark covering diverse editing scenarios and claims consistent outperformance over baselines on instruction fidelity, visual realism, and resistance to forgetting.

Significance. If the central claims hold, the work would be significant as the first dedicated benchmark and method for continual parameter-efficient adaptation in image editing, addressing a practical gap in generative model deployment; the introduction of CIE-Bench as a standardized, difficulty-balanced evaluation protocol is a clear strength that could enable reproducible progress.

major comments (3)
  1. [Abstract and §3] Abstract and §3 (method description): the claim that Adaptive Orthogonal Decoupling 'identifies and orthogonalizes task interference' such that prior-task performance remains intact rests on the unverified assumption that LoRA-matrix orthogonality preserves functional independence under the highly non-linear diffusion mapping from text+image conditioning to output pixels; no derivation or experiment is shown demonstrating that parameter-space orthogonality implies output-space independence on the image manifold.
  2. [§4] §4 (experiments): the abstract asserts outperformance on instruction fidelity, realism, and forgetting robustness, yet the provided text contains no quantitative tables, ablation results, or details on controls (e.g., how CIE-Bench tasks are sequenced, what metrics quantify 'robustness to forgetting'); without these the central empirical claim cannot be assessed.
  3. [§3.2] §3.2 (Rank-Invariant Historical Information Compression): the scalability strategy is described only at a high level; it is unclear whether the compression preserves the orthogonality constraints enforced by Adaptive Orthogonal Decoupling or introduces new interference, which is load-bearing for the continual-learning guarantee.
minor comments (2)
  1. [§3] Notation for the orthogonality constraint and the rank-invariant compression operator should be defined explicitly with equations rather than prose descriptions.
  2. [§4.1] CIE-Bench task descriptions and difficulty balancing criteria are mentioned but not enumerated; a table listing the editing operations, prompt styles, and dataset sources would improve clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important areas for improving theoretical clarity, experimental presentation, and methodological detail. We respond to each major comment below and indicate planned revisions.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3 (method description): the claim that Adaptive Orthogonal Decoupling 'identifies and orthogonalizes task interference' such that prior-task performance remains intact rests on the unverified assumption that LoRA-matrix orthogonality preserves functional independence under the highly non-linear diffusion mapping from text+image conditioning to output pixels; no derivation or experiment is shown demonstrating that parameter-space orthogonality implies output-space independence on the image manifold.

    Authors: We acknowledge that the manuscript does not provide a formal derivation connecting LoRA parameter orthogonality to functional independence in the non-linear diffusion output space. The approach is motivated by reducing interference in parameter space, with empirical support from CIE-Bench results showing preserved prior-task performance. In revision we will add a dedicated discussion subsection and an ablation experiment that measures output-space similarity (e.g., via perceptual metrics) before and after orthogonalization to better substantiate the assumption. revision: partial

  2. Referee: [§4] §4 (experiments): the abstract asserts outperformance on instruction fidelity, realism, and forgetting robustness, yet the provided text contains no quantitative tables, ablation results, or details on controls (e.g., how CIE-Bench tasks are sequenced, what metrics quantify 'robustness to forgetting'); without these the central empirical claim cannot be assessed.

    Authors: The full manuscript contains §4 with quantitative tables, ablation studies, task sequencing details for CIE-Bench, and forgetting metrics (performance retention on prior tasks). We apologize if these elements were not visible in the reviewed version and will ensure all tables, controls, and metric definitions are explicitly presented and cross-referenced in the revised submission. revision: yes

  3. Referee: [§3.2] §3.2 (Rank-Invariant Historical Information Compression): the scalability strategy is described only at a high level; it is unclear whether the compression preserves the orthogonality constraints enforced by Adaptive Orthogonal Decoupling or introduces new interference, which is load-bearing for the continual-learning guarantee.

    Authors: We will expand §3.2 with a detailed algorithmic description and analysis showing that the rank-invariant compression operates on the orthogonal subspaces without altering their mutual orthogonality, thereby preserving the interference-mitigation property. A short proof sketch and pseudocode will be added to demonstrate that no new interference is introduced. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper proposes a new empirical method (ACE-LoRA) consisting of Adaptive Orthogonal Decoupling and Rank-Invariant Historical Information Compression, plus a new benchmark (CIE-Bench). No equations, parameter fits presented as predictions, or load-bearing self-citations appear in the provided text. Claims rest on experimental results rather than a derivation that reduces to its own inputs by construction. This matches the common case of a self-contained method paper with independent empirical content.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; assessment limited to high-level claims.

pith-pipeline@v0.9.1-grok · 5727 in / 1000 out tokens · 21251 ms · 2026-06-30T21:43:18.072476+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

56 extracted references · 12 canonical work pages · 8 internal anchors

  1. [1]

    Instructpix2pix: Learning to follow image editing instructions

    Tim Brooks, Aleksander Holynski, and Alexei A Efros. Instructpix2pix: Learning to follow image editing instructions. InCVPR, pages 18392–18402, 2023

    2023

  2. [2]

    Continual learning with tiny episodic memories

    Arslan Chaudhry, Marcus Rohrbach, Mohamed Elhoseiny, Thalaiyasingam Ajanthan, P Dokania, P Torr, and M Ranzato. Continual learning with tiny episodic memories. InWorkshop on Multi-Task and Lifelong Reinforcement Learning, 2019

    2019

  3. [3]

    Coin: A benchmark of continual instruction tuning for multimodel large language models.NeurIPS, 2024

    Cheng Chen, Junchen Zhu, Xu Luo, Heng T Shen, Jingkuan Song, and Lianli Gao. Coin: A benchmark of continual instruction tuning for multimodel large language models.NeurIPS, 2024

    2024

  4. [4]

    Sefe: Superficial and essential forgetting eliminator for multimodal continual instruction tuning

    Jinpeng Chen, Runmin Cong, Yuzhi Zhao, Hongzheng Yang, Guangneng Hu, Horace Ho Shing Ip, and Sam Kwong. Sefe: Superficial and essential forgetting eliminator for multimodal continual instruction tuning. 2025

    2025

  5. [5]

    Adapt- former: Adapting vision transformers for scalable visual recognition.NeurIPS, 2022

    Shoufa Chen, Chongjian Ge, Zhan Tong, Jiangliu Wang, Yibing Song, Jue Wang, and Ping Luo. Adapt- former: Adapting vision transformers for scalable visual recognition.NeurIPS, 2022

    2022

  6. [6]

    Diffedit: Diffusion-based semantic image editing with mask guidance.arXiv preprint arXiv:2210.11427, 2022

    Guillaume Couairon, Jakob Verbeek, Holger Schwenk, and Matthieu Cord. Diffedit: Diffusion-based semantic image editing with mask guidance.arXiv preprint arXiv:2210.11427, 2022

    work page arXiv 2022

  7. [7]

    Diffusion models beat gans on image synthesis.NeurIPS, 2021

    Prafulla Dhariwal and Alexander Nichol. Diffusion models beat gans on image synthesis.NeurIPS, 2021

    2021

  8. [8]

    Scaling rectified flow transformers for high-resolution image synthesis

    Patrick Esser, Sumith Kulal, Andreas Blattmann, Rahim Entezari, Jonas Müller, Harry Saini, Yam Levi, Dominik Lorenz, Axel Sauer, Frederic Boesel, et al. Scaling rectified flow transformers for high-resolution image synthesis. InICML, 2024

    2024

  9. [9]

    Orthogonal gradient descent for continual learning

    Mehrdad Farajtabar, Navid Azizan, Alex Mott, and Ang Li. Orthogonal gradient descent for continual learning. InInternational conference on artificial intelligence and statistics, 2020

    2020

  10. [10]

    Catastrophic forgetting in connectionist networks.Trends in cognitive sciences, 1999

    Robert M French. Catastrophic forgetting in connectionist networks.Trends in cognitive sciences, 1999

    1999

  11. [11]

    Cyclical Annealing Schedule: A Simple Approach to Mitigating KL Vanishing

    Hao Fu, Chunyuan Li, Xiaodong Liu, Jianfeng Gao, Asli Celikyilmaz, and Lawrence Carin. Cyclical annealing schedule: A simple approach to mitigating kl vanishing.arXiv preprint arXiv:1903.10145, 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  12. [12]

    Ddgr: Continual learning with deep diffusion-based generative replay

    Rui Gao and Weiwei Liu. Ddgr: Continual learning with deep diffusion-based generative replay. InICML, 2023

    2023

  13. [13]

    Hide- llava: Hierarchical decoupling for continual instruction tun- ing of multimodal large language model.arXiv preprint arXiv:2503.12941, 2025

    Haiyang Guo, Fanhu Zeng, Ziwei Xiang, Fei Zhu, Da-Han Wang, Xu-Yao Zhang, and Cheng-Lin Liu. Hide-llava: Hierarchical decoupling for continual instruction tuning of multimodal large language model. arXiv preprint arXiv:2503.12941, 2025

    work page arXiv 2025

  14. [14]

    Lora+: Efficient low rank adaptation of large models

    Soufiane Hayou, Nikhil Ghosh, and Bin Yu. Lora+: Efficient low rank adaptation of large models. 2024

    2024

  15. [15]

    Prompt-to-Prompt Image Editing with Cross Attention Control

    Amir Hertz, Ron Mokady, Jay Tenenbaum, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Prompt-to- prompt image editing with cross attention control.arXiv preprint arXiv:2208.01626, 2022

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  16. [16]

    Denoising diffusion probabilistic models.NeurIPS, 2020

    Jonathan Ho, Ajay Jain, and Pieter Abbeel. Denoising diffusion probabilistic models.NeurIPS, 2020

    2020

  17. [17]

    Parameter-efficient transfer learning for nlp

    Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. Parameter-efficient transfer learning for nlp. InICML, 2019

    2019

  18. [18]

    Lora: Low-rank adaptation of large language models.ICLR, 2022

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models.ICLR, 2022

    2022

  19. [19]

    Cl-moe: Enhancing multimodal large language model with dual momentum mixture-of-experts for continual visual question answering

    Tianyu Huai, Jie Zhou, Xingjiao Wu, Qin Chen, Qingchun Bai, Ze Zhou, and Liang He. Cl-moe: Enhancing multimodal large language model with dual momentum mixture-of-experts for continual visual question answering. InCVPR, 2025

    2025

  20. [20]

    T2i-conbench: Text-to-image benchmark for continual post-training

    Zhehao Huang, Yuhang Liu, Yixin Lou, Zhengbao He, Mingzhen He, Wenxing Zhou, Tao Li, Kehan Li, Zeyi Huang, and Xiaolin Huang. T2i-conbench: Text-to-image benchmark for continual post-training. arXiv preprint arXiv:2505.16875, 2025. 10

    work page arXiv 2025

  21. [21]

    Visual prompt tuning

    Menglin Jia, Luming Tang, Bor-Chun Chen, Claire Cardie, Serge Belongie, Bharath Hariharan, and Ser-Nam Lim. Visual prompt tuning. InECCV, 2022

    2022

  22. [22]

    Overcoming catastrophic forgetting in neural networks.Proceedings of the national academy of sciences, 2017

    James Kirkpatrick, Razvan Pascanu, Neil Rabinowitz, Joel Veness, Guillaume Desjardins, Andrei A Rusu, Kieran Milan, John Quan, Tiago Ramalho, Agnieszka Grabska-Barwinska, et al. Overcoming catastrophic forgetting in neural networks.Proceedings of the national academy of sciences, 2017

    2017

  23. [23]

    Viescore: Towards explainable metrics for conditional image synthesis evaluation

    Max Ku, Dongfu Jiang, Cong Wei, Xiang Yue, and Wenhu Chen. Viescore: Towards explainable metrics for conditional image synthesis evaluation. InACL, 2024

    2024

  24. [24]

    Multi-concept customization of text-to-image diffusion

    Nupur Kumari, Bingliang Zhang, Richard Zhang, Eli Shechtman, and Jun-Yan Zhu. Multi-concept customization of text-to-image diffusion. InCVPR, 2023

    2023

  25. [25]

    FLUX.2: Frontier Visual Intelligence.https://bfl.ai/blog/flux-2, 2025

    Black Forest Labs. FLUX.2: Frontier Visual Intelligence.https://bfl.ai/blog/flux-2, 2025

    2025

  26. [26]

    Black Forest Labs, Stephen Batifol, Andreas Blattmann, Frederic Boesel, Saksham Consul, Cyril Diagne, Tim Dockhorn, Jack English, Zion English, Patrick Esser, et al. Flux. 1 kontext: Flow matching for in-context image generation and editing in latent space.arXiv preprint arXiv:2506.15742, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  27. [27]

    The Power of Scale for Parameter-Efficient Prompt Tuning

    Brian Lester, Rami Al-Rfou, and Noah Constant. The power of scale for parameter-efficient prompt tuning. arXiv preprint arXiv:2104.08691, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  28. [28]

    Prefix-Tuning: Optimizing Continuous Prompts for Generation

    Xiang Lisa Li and Percy Liang. Prefix-tuning: Optimizing continuous prompts for generation.arXiv preprint arXiv:2101.00190, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  29. [29]

    Learning without forgetting.TPAMI, 2017

    Zhizhong Li and Derek Hoiem. Learning without forgetting.TPAMI, 2017

    2017

  30. [30]

    Pcr: Proxy-based contrastive replay for online class-incremental continual learning

    Huiwei Lin, Baoquan Zhang, Shanshan Feng, Xutao Li, and Yunming Ye. Pcr: Proxy-based contrastive replay for online class-incremental continual learning. InCVPR, 2023

    2023

  31. [31]

    Keeplora: Continual learning with residual gradient adaptation

    Mao-Lin Luo, Zi-Hao Zhou, Yi-Lin Zhang, Yuanyu Wan, Tong Wei, and Min-Ling Zhang. Keeplora: Continual learning with residual gradient adaptation. 2026

    2026

  32. [32]

    Catastrophic interference in connectionist networks: The sequential learning problem

    Michael McCloskey and Neal J Cohen. Catastrophic interference in connectionist networks: The sequential learning problem. InPsychology of learning and motivation. 1989

    1989

  33. [33]

    SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations

    Chenlin Meng, Yutong He, Yang Song, Jiaming Song, Jiajun Wu, Jun-Yan Zhu, and Stefano Ermon. Sdedit: Guided image synthesis and editing with stochastic differential equations.arXiv preprint arXiv:2108.01073, 2021

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  34. [34]

    Null-text inversion for editing real images using guided diffusion models

    Ron Mokady, Amir Hertz, Kfir Aberman, Yael Pritch, and Daniel Cohen-Or. Null-text inversion for editing real images using guided diffusion models. InCVPR, 2023

    2023

  35. [35]

    Continual lifelong learning with neural networks: A review.Neural networks, 2019

    German I Parisi, Ronald Kemker, Jose L Part, Christopher Kanan, and Stefan Wermter. Continual lifelong learning with neural networks: A review.Neural networks, 2019

    2019

  36. [36]

    Scalable diffusion models with transformers

    William Peebles and Saining Xie. Scalable diffusion models with transformers. InICCV, 2023

    2023

  37. [37]

    SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis

    Dustin Podell, Zion English, Kyle Lacey, Andreas Blattmann, Tim Dockhorn, Jonas Müller, Joe Penna, and Robin Rombach. Sdxl: Improving latent diffusion models for high-resolution image synthesis.arXiv preprint arXiv:2307.01952, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  38. [38]

    Correlated low-rank adaptation for convnets

    Wu Ran, Weijia Zhang, ShuYang Pang, Qi Zhu, Jinfan Liu, JingSheng Liu, Xin Cao, Qiang Li, Yichao Yan, and Chao Ma. Correlated low-rank adaptation for convnets. InNeurIPS, 2025

    2025

  39. [39]

    Experience replay for continual learning

    David Rolnick, Arun Ahuja, Jonathan Schwarz, Timothy Lillicrap, and Gregory Wayne. Experience replay for continual learning. InNeurIPS, 2019

    2019

  40. [40]

    High-resolution image synthesis with latent diffusion models

    Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. InCVPR, 2022

    2022

  41. [41]

    Lfs-gan: Lifelong few-shot image generation

    Juwon Seo, Ji-Su Kang, and Gyeong-Moon Park. Lfs-gan: Lifelong few-shot image generation. InICCV, 2023

    2023

  42. [42]

    Coda-prompt: Continual decomposed attention- based prompting for rehearsal-free continual learning

    James Seale Smith, Leonid Karlinsky, Vyshnavi Gutta, Paola Cascante-Bonilla, Donghyun Kim, Assaf Arbelle, Rameswar Panda, Rogerio Feris, and Zsolt Kira. Coda-prompt: Continual decomposed attention- based prompting for rehearsal-free continual learning. InCVPR, 2023

    2023

  43. [43]

    Adaptive memory replay for continual learning

    James Seale Smith, Lazar Valkov, Shaunak Halbe, Vyshnavi Gutta, Rogerio Feris, Zsolt Kira, and Leonid Karlinsky. Adaptive memory replay for continual learning. InCVPR, 2024. 11

    2024

  44. [44]

    Model merging with svd to tie the knots.arXiv preprint arXiv:2410.19735, 2024

    George Stoica, Pratik Ramesh, Boglarka Ecsedi, Leshem Choshen, and Judy Hoffman. Model merging with svd to tie the knots.arXiv preprint arXiv:2410.19735, 2024

    work page arXiv 2024

  45. [45]

    Lora merging with svd: Understanding interference and preserving performance

    Dennis Tang, Prateek Yadav, Yi-Lin Sung, Jaehong Yoon, and Mohit Bansal. Lora merging with svd: Understanding interference and preserving performance. InICML, 2025

    2025

  46. [46]

    Hydralora: An asymmetric lora architecture for efficient fine-tuning

    Chunlin Tian, Zhan Shi, Zhijiang Guo, Li Li, and Chengzhong Xu. Hydralora: An asymmetric lora architecture for efficient fine-tuning. 2024

    2024

  47. [47]

    Orthogonal subspace learning for language model continual learning

    Xiao Wang, Tianze Chen, Qiming Ge, Han Xia, Rong Bao, Rui Zheng, Qi Zhang, Tao Gui, and Xuan-Jing Huang. Orthogonal subspace learning for language model continual learning. InFindings of the Association for Computational Linguistics: EMNLP, 2023

    2023

  48. [48]

    Learning to prompt for continual learning

    Zifeng Wang, Zizhao Zhang, Chen-Yu Lee, Han Zhang, Ruoxi Sun, Xiaoqi Ren, Guolong Su, Vincent Perot, Jennifer Dy, and Tomas Pfister. Learning to prompt for continual learning. InCVPR, 2022

    2022

  49. [49]

    Smolora: Exploring and defying dual catastrophic forgetting in continual visual instruction tuning

    Ziqi Wang, Chang Che, Qi Wang, Yangyang Li, Zenglin Shi, and Meng Wang. Smolora: Exploring and defying dual catastrophic forgetting in continual visual instruction tuning. InCVPR, 2025

    2025

  50. [50]

    Ties-merging: Resolving interference when merging models

    Prateek Yadav, Derek Tam, Leshem Choshen, Colin A Raffel, and Mohit Bansal. Ties-merging: Resolving interference when merging models. 2023

    2023

  51. [51]

    Qwen3 Technical Report

    An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. Qwen3 technical report.arXiv preprint arXiv:2505.09388, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  52. [52]

    Imgedit: A unified image editing dataset and benchmark

    Yang Ye, Xianyi He, Zongjian Li, Bin Lin, Shenghai Yuan, Zhiyuan Yan, Bohan Hou, and Li Yuan. Imgedit: A unified image editing dataset and benchmark. 2025

    2025

  53. [53]

    Boosting continual learning of vision-language models via mixture-of-experts adapters

    Jiazuo Yu, Yunzhi Zhuge, Lu Zhang, Ping Hu, Dong Wang, Huchuan Lu, and You He. Boosting continual learning of vision-language models via mixture-of-experts adapters. InCVPR, 2024

    2024

  54. [54]

    Language models are super mario: Absorbing abilities from homologous models as a free lunch

    Le Yu, Bowen Yu, Haiyang Yu, Fei Huang, and Yongbin Li. Language models are super mario: Absorbing abilities from homologous models as a free lunch. InICML, 2024

    2024

  55. [55]

    Lifelong gan: Continual learning for conditional image generation

    Mengyao Zhai, Lei Chen, Frederick Tung, Jiawei He, Megha Nawhal, and Greg Mori. Lifelong gan: Continual learning for conditional image generation. InICCV, 2019

    2019

  56. [56]

    Bilora: Almost-orthogonal parameter spaces for continual learning

    Hao Zhu, Yifei Zhang, Junhao Dong, and Piotr Koniusz. Bilora: Almost-orthogonal parameter spaces for continual learning. InCVPR, 2025. 12 Appendix A Related Work Diffusion-Based Image Editing.Large-scale diffusion models [ 40, 37, 8, 26, 36] have demonstrated remarkable success in synthesizing high-fidelity and semantically complex images from textual pro...

    2025