No.26-05 No-Regret Self-Improvement: A Unified Game-Theoretic Formulation for AI Self-Improvement
Ensuring that AI systems improve consistently under iterative feedback is a central challenge in AI self-improvement. This talk prese...
No.26-03 OpenClaw-RL: Train Any Agent Simply by Talking
From LLM, Mar 31, 2026
No.26-02 Simulating Real Users with State Alignment
From LLM, Mar 24, 2026All Talks
No.26-05 No-Regret Self-Improvement: A Unified Game-Theoretic Formulation for AI Self-Improvement
Ensuring that AI systems improve consistently under iterative feedback is a central challenge in AI self-improvement. This talk presents a unified game-theoretic formulation that conn...
From LLM, Oct 07, 2026
No.26-04 Beyond the Surface: How Post-Training Artifacts Shape LLM Diversity and Safety
Post-training alignment makes LLMs helpful, but also introduces unintended artifacts. This talk explores two such artifacts, their impact on LLM diversity and safety, and presents cor...
From LLM, Apr 08, 2026
No.26-03 OpenClaw-RL: Train Any Agent Simply by Talking
OpenClaw-RL is an RL server that allows users to deploy their personal models and continuously improve them through everyday interactions in OpenClaw. We propose an optimization metho...
From LLM, Mar 31, 2026
No.26-02 Simulating Real Users with State Alignment
User simulation offers a scalable path toward more human-centered AI systems. However, current approaches largely rely on response imitation, capturing surface-level language patterns...
From LLM, Mar 24, 2026
No.26-01 Structured Representation Learning for Latent Thinking in LLMs
Despite the excellent performance of LLMs across various output-based benchmarks, we argue that focusing solely on outputs overlooks both underlying risks and untapped opportunities w...
From LLM, Feb 11, 2026
Beyond Visual Geometry- Toward Physical 3D Reconstruction
Recent progress in 3D Visual Geometry has led to impressive reconstruction and generation results using purely feed-forward deep networks. However, much of this progress remains limit...
From NTU, Sep 26, 2025
AI-Driven Solutions in Individualised Medicine- From Multimodal Omics Data to Disease Diagnosis and Biomarker Discovery
Despite significant advancements in medical science and an increasing emphasis on precision medicine, the majority of medical diagnoses are still made after patients exhibit noticeabl...
From UNSW, Jul 18, 2025
Teach AI What It Doesn't Know
The remarkable capabilities of machine learning (ML) models, especially foundation models like GPT, have transformed numerous domains. However, these systems often falter in real-worl...
From NTU, Jun 26, 2025
Reasoning with Language Models
Language models are primarily trained via imitation on massive amounts of human data; as a result, they’re capable of performing a wide range of tasks but often lack the deep reasonin...
From UC Berkeley, May 29, 2025
Towards Efficient Novel View Synthesis
Novel view synthesis (NVS) from 2D images aims to generate unseen views of a scene given multiple input observations. It is a fundamental task in computer vision that has garnered sig...
From Monash University, May 23, 2025