Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors
本文提出Tactile-JEPA，一种基于触觉传感器空间布局的自监督预训练方法，以提高分布式触觉传感器表示学习的质量，从而改善机器人操作性能。
0 citationsRead paper
Selected work
Representative Papers
What Attention Recalls and Recurrence Controls in Hybrid Language Models
研究通过两种干预方法探讨了混合语言模型中注意力和循环状态的角色，发现注意力负责精确检索，而循环状态控制输出语言和个性。
0 citationsRead paper
Enoki: Efficient Multi-Level Hallucination Detection
为解决LLM在高风险场景中的事实性问题，提出Enoki框架，通过多级幻觉检测方法，结合文本锚定关系事实的提取与验证，实现高效的事实核验和定位。
0 citationsRead paper
MemToC: Benchmarking Memory-Tool Conflict Resolution in Large Language Models
研究解决了工具增强型大语言模型在工具返回与参数记忆冲突时的仲裁问题，通过引入MemToC基准进行评估，并采用SFT和DPO方法改进了正确答案保留率。
0 citationsRead paper
The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearning
This study addresses the limitation of existing large language model unlearning methods that overlook fact popularity, rendering high-frequency knowledge difficult to remove. We propose AdaPop, a novel approach that models popularity as a learnable parameter by integrating token confidence with external proxy evaluations. Through a bi-ascent controller, AdaPop dynamically adjusts penalty intensity to achieve an adaptive balance between forgetting and retention. Experiments across three model families and two benchmarks demonstrate that AdaPop reduces content leakage under paraphrased queries by approximately fivefold and under adversarial reconstruction by 1.6 times. Furthermore, internal representation analysis confirms superior forgetting separation, effectively resolving the challenge of unlearning high-frequency facts while preserving general model utility.
0 citationsRead paper
Recent publications
Latest Papers
Tactile-JEPA: Topology-Aware Self-Supervised Representation Learning for Distributed Tactile Sensors
本文提出Tactile-JEPA，一种基于触觉传感器空间布局的自监督预训练方法，以提高分布式触觉传感器表示学习的质量，从而改善机器人操作性能。
0 citationsRead paper
What Attention Recalls and Recurrence Controls in Hybrid Language Models
研究通过两种干预方法探讨了混合语言模型中注意力和循环状态的角色，发现注意力负责精确检索，而循环状态控制输出语言和个性。
0 citationsRead paper
Enoki: Efficient Multi-Level Hallucination Detection
为解决LLM在高风险场景中的事实性问题，提出Enoki框架，通过多级幻觉检测方法，结合文本锚定关系事实的提取与验证，实现高效的事实核验和定位。
0 citationsRead paper
MemToC: Benchmarking Memory-Tool Conflict Resolution in Large Language Models
研究解决了工具增强型大语言模型在工具返回与参数记忆冲突时的仲裁问题，通过引入MemToC基准进行评估，并采用SFT和DPO方法改进了正确答案保留率。
0 citationsRead paper
The More Popular, The Harder to Forget: Adaptive Popularity for LLM Unlearning
This study addresses the limitation of existing large language model unlearning methods that overlook fact popularity, rendering high-frequency knowledge difficult to remove. We propose AdaPop, a novel approach that models popularity as a learnable parameter by integrating token confidence with external proxy evaluations. Through a bi-ascent controller, AdaPop dynamically adjusts penalty intensity to achieve an adaptive balance between forgetting and retention. Experiments across three model families and two benchmarks demonstrate that AdaPop reduces content leakage under paraphrased queries by approximately fivefold and under adversarial reconstruction by 1.6 times. Furthermore, internal representation analysis confirms superior forgetting separation, effectively resolving the challenge of unlearning high-frequency facts while preserving general model utility.
0 citationsRead paper