Zeguan Xiao, Zhuocheng Liang, Yun Chen, Yong Wang, Guanhua Chen, Pinyan Lu · Curvature-Constrained Adaptation
CurvDelta: Low-Curvature Delta Modeling For Surgical Learning and Unlearning
CurvDelta is a framework that builds capability preservation into the parameterisation of LLM updates, using a low-curvature space around the pre-adaptation model, approximated with K-FAC and a factorised projection. The authors also introduce CurvLoRA, a low-rank variant for more efficient adaptation. The paper reports competitive maths reasoning and code generation with better retention of general capabilities, and a better forgetting-retention balance than baselines on TOFU and MUSE for knowledge unlearning.
Open the source ↗Similar Research & Safety releases
- Test-time adaptation for AI text detectors
MIT and Cornell Tech researchers (Kevin Ren, Manish Raghavan, Nikhil Garg) · Adaptive AI Detector · Week 48, 2026
- LightTAR: Light and Fast Target-Aware Representations for Multimodal Tabular Learning
Eilam Shapira, Alan Arazi, Shoham Grunblat, Roi Reichart (Technion) · Target-aware tabular encoding · Week 48, 2026
- Self-Improving Memory
Meta FAIR · Self-Improving Memory · Week 48, 2026
- TuringSpace: Grounding Spatial Intelligence in Action with Interactive CAPTCHAs
Meta (Xuwei Ding et al.) · Spatial Action Benchmark · Week 48, 2026
- WeVisForensics: Self-Evolving Agent for Omni AI-Generated Video Detection
Tencent (Yipeng Du, Yuxi Li, Ying Zhang, Chong Sun, Chen Li, Jing Lyu, Ying Tai) · Agentic Video Forensics · Week 48, 2026
- Predicting End-to-End Fine-Tuning Time for Generative Language Models
Dharma-AI (Hugging Face blog) · Fine-tuning Time Predictor · Week 48, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.