cs.CVNov 19, 2026

HOPD: Historical On-Policy Distillation for SAR MLLM Continual Learning

Authors: Yuxuan Li, Derrick Goh Xin Deik, Zheng-Peng Duan, Qianru Sun, Weijie Li

Organizations: Singapore Management University · Nanyang Technological University · Nankai University · Xi’an Jiaotong University

Abstract

Continual SAR interpretation requires multimodal large language models to learn incoming sources while retaining earlier task abilities. This is difficult when acquisition and geographic shifts alter visual evidence, regionally concentrated categories bias answers, and historical images are scarce. We address this problem with Historical On-Policy Distillation (HOPD), which reuses source-acquisition LoRA states as persistent teachers for one evolving student. Archiving each state before consolidation preserves a stable source reference. The teachers supervise student-generated prefixes, transferring historical behavior to the student's changing generation contexts. To enrich sparse replay, original-view teachers guide augmented historical student views while preserving spatial coordinates. This knowledge is consolidated into a single deployed adapter that requires no source identity. On SARVQA-CL, our source-incremental evaluation protocol, HOPD achieves the highest final mean score of 54.51 among the compared methods, exceeding the strongest baseline by 2.96 points. It also improves the final source-macro score in every evaluated task over shared-memory Replay-SFT, with gains across source orders. These results support source-acquisition history as supervision for retaining earlier task performance during continual SAR adaptation. Code and data: https://github.com/zcablii/HOPD.

Figures & tables

Appendix figures & tables7 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Jun 22, 2026

Attention-Spectrum Regularization for Replay-Free Continual Multimodal LLMs

Multimodal large language models (MLLMs) are increasingly required to adapt to non-stationary streams of visual domains, question types, and user instructions, yet continual fine-tuning often causes severe forgetting of previously acquired multimodal skills. Existing continual vision-language methods mainly preserve outputs, replay data or pseudo-data, regularize embedding geometry, or allocate task-specific parameters, but they provide limited control over how internal cross-modal attention patterns supporting old skills drift during adaptation. We propose Attention-Spectrum Regularization (ASR), a replay-free continual learning framework that preserves skill-conditioned structures of cross-modal attention. ASR treats cross-attention maps as two-dimensional signals, summarizes their scale and directional properties into compact spectral statistics, and stores only skill-wise prototype distributions instead of replaying past image-question pairs, generated pseudo-examples, or old-stage teacher snapshots. In later stages, a phase-invariant spectral regularizer constrains harmful drift of these prototypes while allowing instance-level attention to adapt to new tasks. We provide theoretical analysis showing that skill-conditioned spectral drift controls forgetting under a spectral sufficiency assumption, and that Fourier power spectra are stable to spatial translations and bounded perturbations. Experiments on continual VQA and multimodal instruction-tuning benchmarks, including VQA v2, VQACL, CLT-VQA, CoIN, and UCIT, show that ASR consistently improves final performance and reduces forgetting over strong replay-, regularization-, and adapter-based baselines. Preserving skill-level attention structure is an effective and lightweight mechanism for continual MLLMs. Code is available at https://github.com/Creative-zcx/attention-spectrum-replay
May 28, 2026

On-Policy Replay for Continual Supervised Fine-Tuning

Continual supervised fine-tuning (SFT) is the de facto recipe for adapting large language models (LLMs) to a stream of downstream tasks, but it suffers from catastrophic forgetting of earlier capabilities. Recent work shows that on-policy signals -- training on the model's own outputs -- reduce forgetting more reliably than off-policy supervision. Existing on-policy methods route this signal through a new training objective (e.g., self-distillation losses with a teacher copy), inheriting an extra forward pass, schedule sensitivity, and stylistic drift from the teacher.We instead route the on-policy signal through the training data source. Our method, On-Policy Replay (OPR), rolls out the most recent checkpoint on a small budget of historical prompts, filters the generations by a task reward, and replays the surviving (prompt, model response) pairs as ordinary SFT examples. There is no teacher, no auxiliary loss, and no on-the-fly distillation. Across three 7--8B instruction-tuned backbones (Qwen2.5-7B-Instruct, Qwen3-8B, Llama3.1-8B-Instruct) on the TRACE continual-learning benchmark, OPR consistently reduces forgetting; on the sharpest stress test (Qwen2.5-7B-Instruct, Sequential SFT BWT -13.93), OPR lifts BWT to -0.65 at a 10% replay budget and to -2.29 at a 1% budget -- a 46% reduction in |BWT| over a tuned Vanilla Replay baseline, with 42--46% reductions observed across all three backbones. We give a KL-shrinkage interpretation that places OPR and prior on-policy distillation methods on a single axis, and we present a counterintuitive finding that explains why Vanilla Replay is already a strong baseline: low-score replay is uniformly worse than Vanilla Replay, demonstrating that the active ingredient in OPR is the on-policy distribution, not the response quality alone.Our code is available at https://github.com/Yancey2024/OnPolicyReplay.
Oct 27, 2026

CAMR: Compatibility-Aware Memory Replay for Continual Learning in Large Language Models

Continual learning requires large language models to acquire new knowledge and capabilities while retaining what they have learned. Fine-tuning and replay support acquisition and retention, but deciding whether incoming examples need supervision and which historical examples to revisit remains a challenge. We propose Compatibility-Aware Memory Replay (CAMR), an adaptive framework that jointly controls current-example supervision and historical replay. CAMR combines compatibility checks with a learned memory gate to guide an online policy over three actions: learning incoming examples with general replay, jointly reinforcing compatible related examples, and skipping direct supervision of verified mastered duplicates. The policy learns from acquisition, retention, and training-cost feedback under a bounded per-arrival replay budget. We evaluate CAMR with Qwen3-8B-Base on continual factual updating and cross-task and cross-domain learning. On WikiBigEdit, CAMR consistently leads evaluated methods in Original Match from 3K to 100K facts, reaching 93.66% after just 3K facts and surpassing Full-history Replay (LoRA) after 100K facts. It also achieves the highest final scores on all four selected TRACE tasks and maintains or improves earlier-task performance relative to post-acquisition levels in a sequence spanning financial question answering, mathematical reasoning, code generation, and summarization.