Continual harness evolution causes forgetting in coding agents
Meta-Harness drops 66% on SWE-Bench after evolution on FeatureBench with DeepSeek-V4-Flash-0731.
Stories per day · 30 days
2 RSS
Relatedcontinual-learning2agent-harness1asr1coding-agents1concept-learning1continual-evolution1data-weighting1explainability1fine-tuning1model-merging1multilingual1saliency1self-evolution1shapley1speech-llm1
15 stories, newest first · one card per story
Stories tagged #catastrophic-forgetting
Speech LLM ASR expansion forgets source languages unless merges are selective
SSU-Wanda+TIES best preserved source languages in single-target expansion; recursive multi-target updates were highly order-sensitive.
Concept-consistent continual learning stops saliency explanation drift
C3L anchors regularization to offline class concepts, and authors report gains on five benchmarks against over 20 baselines.
DRIVE weights fine-tuning samples to limit forgetting without source data
Per-sample weights combine Shapley anti-forgetting and adaptation signals with diversity and a prior.
Self-Preservation in Language Models
Qwen models can generate replay and sometimes choose tasks that better preserve prior capabilities.
When Forgetting is not Catastrophic: On the Mechanics of Spurious Forgetting
Spurious forgetting hides old facts through a reversible shared shift before slower fact-specific erosion.
PaLoRA: Paced Low-Rank Adaptation for Continual Learning
PaLoRA paces LoRA continual-learning steps by effective rank, reducing forgetting on long task sequences.
What Should World Models Forget? Stratified Retention for Continual Adaptation
Paper proposes stratified retention for continual world models, separating never-revisable invariants like physics from instance-level facts, plus a differential retention metric.
Local Support Learning
Local Support Learning limits catastrophic forgetting in LLMs up to 7B without replaying prior data.
No Free Efficiency: Revisiting the Trade-off Between Training Efficiency and Model Vulnerability
Efficient training recipes make vision and language models easier to attack and less robust.
When Can First-Order Models of Fine-Tuning Bound Forgetting?
First-order LoRA probes bound forgetting only when response coefficients stay within the margin distance.
Fine-Tuning LLMs for Translation: General Forgetting Mitigation Does Not Preserve MT-Specific Instruction Following
Elastic Weight Consolidation limits general forgetting in translation fine-tuning but not translation-specific instruction following.
Navigating the Latent Manifold: Proactive Concept Drift Adaptation for Resilient NIDS
Researchers propose DriftXpert, a concept-drift-adaptive network intrusion detection system validated on enterprise networks, addressing degraded AI-based NIDS performance in dynamic traffic.
Qwen-Drive 1.0 tells you why it brakes, just don't expect the explanation to match the maneuver
Alibaba's Qwen-Drive 1.0 adds 3D perception and planning modules to Qwen3.5-4B for driving tasks, though explanations often mismatch maneuvers.
2
Continual Learning Mechanisms Compose for Long-Horizon Memorization
Composing data, function, and weight anchors with merged LoRA raises 100-task long-horizon retention from 1.2% to 34.9% in continual fine-tuning.