Recent advances enable LLMs to generate simulation code from natural language, yet interpreting 3D physical field outputs remains unsolved. Existing 3D scene compression methods fail on physical fields due to absent semantic grounding and information loss. We discover that typical physical fields exhibit extreme information redundancy, motivating a paradigm shift from lossy compression to selective exploration. Building on this insight, we propose AQUA, which trains agents to actively query information-rich regions through Gaussian Splatting environments, transforming global lossy compression into local lossless localization with targeted spatial queries. Agents learn query strategies via physics-guided reinforcement learning, overcoming early-stage sparsity without expert-annotated trajectories. On PhysQA-Bench, AQUA achieves 72% accuracy on average, outperforming baselines by 15% across all four datasets. Our code is available at https://github.com/gaoch6258/AQUA.git.
Chonghan Gao, Haoyi Zhou, Zhemeng Luo et al.· Proceedings of the 32nd ACM...· 0 citations
Influence functions provide a principled framework for tracing model predictions back to training data, yet existing methods remain impractical for large language models due to prohibitive storage and I/O costs. Prior approaches face two critical barriers: the Fidelity Gap from coarse curvature approximations, and the I/O bottleneck from materializing per-example gradients, which renders million-scale datasets infeasible. We propose StructInf, a structure-aware influence estimation framework that addresses both gaps through system-algorithm co-design. For fidelity, we first replace the indefinite Hessian with an adaptive block-diagonal damped empirical Fisher surrogate, computed via quadratically convergent Newton--Schulz iterations. For efficiency, we then introduce a streaming algorithm that eliminates per-example gradient storage entirely, reducing space complexity from O(N • d) to O(d) and enabling processing of million-scale training instances on consumer hardware, where N denotes the number of training instances and d the parameter size. We further identify and resolve Split-View Bias, a distributed pitfall where naive averaging of local curvatures fails, via a distributed gradient calibration strategy that synchronizes global statistics with minimal communication. Finally, we conduct comprehensive experiments to evaluate StructInf against other baselines. Unlike prior methods requiring hours of pre-computation or TB-scale storage, StructInf enables real-time data selection during training. Across GLUE and instruction-tuning benchmarks, StructInf achieves up to +26% AUC on MRPC and average 7× speedup for 7B-scale models, making high-fidelity influence analysis practical on a single RTX 4090. These results suggest that practical IF for LoRA-tuned LLMs benefits from joint optimization of curvature modeling, memory management, and distributed coordination.
Mengyi Yan, Yaoshu Wang, Guangyi Zhang et al.· Proceedings of the 32nd ACM...· 0 citations
