Tian-Fan Fu

Cell-Lens: Advancing LLM Reasoning at Single-Cell Resolution

MatToolBench: Benchmarking Multimodal Agents in Real-World Materials Science Workflows

The experiments show that strong performance on general benchmarks does not transfer to professional scientific workflows, and that this gap is not a visual-grounding problem alone: failures arise from domain-specific operational knowledge, sparse pretraining coverage of scientific software, weak cross-tool artifact ha...

Equivariant Neural Primal-Dual Assignment for Maximum Common Edge Subgraphs

This work introduces Equivariant Neural Primal-Dual Assignment (ENPDA), which learns a shared matching policy and applies it to new pairs without further training, answering queries roughly three orders of magnitude faster and recovering its training cost after a few dozen queries.

Disagree to Accelerate: Closing the Loop on Diffusion Feature Forecasts

Training-free feature forecasting accelerates diffusion sampling by predicting features at skipped denoising steps. Recent work has mainly focused on designing stronger forecasters. Yet forecast error varies sharply across steps, and open-loop caches trust the forecast in full at every skipped step. This fixed trust is...

Compute Allocation for Self-Evolving LLMs: From Depth-Breadth to Multi-Armed Bandits

BaSE (Bandit-based Self-Evolving), a multi-armed bandit that allocates LLM calls across parallel trajectories is proposed, a multi-armed bandit that improves mean fitness by 12.3% over the strongest island-protocol baseline across 8 (model, task) cells.

ABOPD: Antibody CDR Design via On-Policy Distillation

ABOPD is introduced, an antibody design framework based on on-policy distillation that leverages privileged native geometry during training to supervise states visited along the model's own denoising trajectories, offering a path to higher-fidelity protein design.

Reasoning BO: Enhancing Bayesian Optimization With the Reasoning Power of LLMs

Reasoning BO is introduced, a novel framework leveraging reasoning models to guide BO sampling while incorporating multi‐agent systems and knowledge graphs for online knowledge accumulation, and it is demonstrated that smaller LLMs, after post‐training, can achieve performance comparable to larger counterparts.

Reachability Is Not Realization: Tracing the Sources of LLM Benchmark Gains

This work establishes a question-level audit under fixed budgets, temperatures, and answer formats, and asks why reachable answers sometimes fail to appear, and test whether inference-time layer routing can expand reachability.

AgentFold: Closed-Loop Agentic Search for Protein Folding Model Design

AgentFold is presented, a multi-agent framework that formulates folding-model development as a closed-loop search over executable code variants and improves the best lDDT by 7.5% over independent Codex proposals and outperforms a random-search control.

Hierarchical Data Selection via Manifold Coverage and Sparse Feature Coverage in LLM Post-training

MASS learns low-dimensional principal manifold coordinates with a dense autoencoder for coarse semantic grouping, and then performs quality-aware sparse feature coverage within each group using a TopK sparse autoencoder and proposes MASS.

Data-DPO: Direct Preference Optimization for Target Model Data Selection in LLM Post-Training

Data-DPO, a target model-oriented SFT data selection method that consistently outperforms existing data selection baselines under multiple data budgets and stably surpasses full data training performance is proposed.