arXiv subjects

Min Lin

Publications and source records attributed to Min Lin.

At least 19 recordsLinked to original sources

arXiv2026

A Backend-Agnostic MWIS Kernel for Stochastic Unit Commitment with Neutral-Atom Hardware Validation

Quantum hardware is beginning to address structured combinatorial optimisation, but two steps still block practical use: mapping real operational models onto hardware-compatible instances, and converting noisy hardware output back into feasible decisions. Here we introduce a backend-agnostic computational interface that compiles the discrete decision layer of stochastic unit commitment into a move-based maximum-weight independent set (MWIS) problem, while retaining continuous dispatch and feasibility recovery in the classical computational layer. We validate the approach in a green hydrogen scheduling setting and deploy it on the QuEra Aquila neutral-atom quantum processor. This is the first end-to-end industrial scheduling workflow that connects real operational decisions to programmable neutral-atom hardware through a solver-agnostic MWIS representation. Across a 15-day hardware campaign on 50-node instances, hardware-generated solutions after classical refinement match or exceed the dispatch margins obtained from exact MWIS on every day. When scaling to 144 nodes, encoding quality remains stable, while the probability that the full atom array survives, rather than graph embedding, emerges as the dominant bottleneck to further scaling. Together, these results establish a hardware-compatible computational pathway toward larger problem scales, and lay the groundwork for exploring regimes in which exact classical optimisation may no longer scale efficiently.

quant-ph↗

arXiv2026

Ai2-Kit: Streamlining AI-Accelerated Ab Initio Workflows for Complex Chemical Systems

Molecular simulations of complex chemical systems, such as catalysis, electrochemistry, and energy storage, often need to capture the interplay of effects such as electronic structure, finite-temperature fluctuations, and electric-field response. Such complexity is difficult to address with traditional ab initio calculations, which are limited by the time and length scales they can reach. AI-accelerated ab initio (AI2) methods use machine learning potentials trained on first-principles data to replace expensive electronic-structure calculations, extending ab initio accuracy to these regimes, but their routine application requires reliable workflows that connect first-principles calculations, model training, molecular dynamics, enhanced sampling, trajectory analysis, and HPC orchestration. Here we present ai2-kit, a software toolkit for developing accessible, reproducible, and extensible AI2 workflows. ai2-kit provides high-semantic-density command-line interfaces and Python APIs for structure and dataset conversion, batch task generation, active-learning screening, job orchestration, and workflow recovery. We demonstrate ai2-kit in four representative applications: active-learning-based machine learning potential construction, free-energy perturbation for redox and acid-base processes, electrochemical machine learning potentials for electrified interfaces, and spectroscopies from machine learning molecular dynamics. ai2-kit also provides AI-agent skills that help users adapt these use cases into customized workflows for their own chemical systems and computational software stacks. Together, ai2-kit helps turn AI2 methods from bespoke computational protocols into reusable and extensible workflows for complex chemical systems, from model construction to property prediction.

physics.chem-ph↗

arXiv2026

Simultaneous confidence bands for cumulative hazard via exchangeable bootstrap and box calibration

Resampling-based simultaneous confidence bands for cumulative hazard functions often undercover in finite samples with right censoring. We study two aspects of the construction that can contribute to this gap, the resampling scheme and the calibration statistic, and propose a procedure that intervenes on both. The exchangeable bootstrap reweights the numerator and the denominator of the Nelson-Aalen ratio, preserving its ratio structure. The box-calibrated discrepancy constructs lower and upper step envelopes from adjacent values of the original and resampled Nelson-Aalen estimators and measures the resulting vertical discrepancy. We establish conditional weak convergence of the exchangeable bootstrap, prove that box calibration is first-order asymptotically equivalent to grid calibration, and show that the resulting band attains nominal coverage asymptotically. The box correction uses the same bootstrap paths and event-time grid as grid calibration; after each bootstrap path is formed, it requires only an additional linear pass over the event-time grid and therefore has negligible computational overhead. In simulations across a range of hazard shapes and censoring levels, the exchangeable bootstrap with box calibration is, in most configurations, closest to nominal coverage among the methods considered. A notable consequence is a ranking reversal: the ratio-preserving exchangeable bootstrap has the lowest coverage under grid calibration, yet is usually closest to the nominal level after box calibration. A melanoma data example illustrates the practical effect on the cumulative hazard bands. The proposed procedure operates on the original cumulative-hazard scale, requires no variance-stabilizing transformation, and permits inference from time zero.

stat.ME↗

arXiv2026

PCSTracker: Long-Term Scene Flow Estimation for Point Cloud Sequences

Point cloud scene flow estimation is fundamental to long-term and fine-grained 3D motion analysis. However, existing methods are typically limited to pairwise settings and struggle to maintain temporal consistency over long sequences as geometry evolves, occlusions emerge, and errors accumulate. In this work, we propose PCSTracker, the first end-to-end framework specifically designed for consistent scene flow estimation in point cloud sequences. Specifically, we introduce an iterative geometry motion joint optimization module (IGMO) that explicitly models the temporal evolution of point features to alleviate correspondence inconsistencies caused by dynamic geometric changes. In addition, a spatio-temporal point trajectory update module (STTU) is proposed to leverage broad temporal context to infer plausible positions for occluded points, ensuring coherent motion estimation. To further handle long sequences, we employ an overlapping sliding-window inference strategy that alternates cross-window propagation and in-window refinement, effectively suppressing error accumulation and maintaining stable long-term motion consistency. Extensive experiments on the synthetic PointOdyssey3D and real-world ADT3D datasets show that PCSTracker achieves the best accuracy in long-term scene flow estimation and maintains real-time performance at 32.5 FPS, while demonstrating superior 3D motion understanding compared to RGB-D-based approaches.

cs.CV↗

arXiv2026

PromptStereo: Zero-Shot Stereo Matching via Structure and Motion Prompts

Modern stereo matching methods have leveraged monocular depth foundation models to achieve superior zero-shot generalization performance. However, most existing methods primarily focus on extracting robust features for cost volume construction or disparity initialization. At the same time, the iterative refinement stage, which is also crucial for zero-shot generalization, remains underexplored. Some methods treat monocular depth priors as guidance for iteration, but conventional GRU-based architectures struggle to exploit them due to the limited representation capacity. In this paper, we propose Prompt Recurrent Unit (PRU), a novel iterative refinement module based on the decoder of monocular depth foundation models. By integrating monocular structure and stereo motion cues as prompts into the decoder, PRU enriches the latent representations of monocular depth foundation models with absolute stereo-scale information while preserving their inherent monocular depth priors. Experiments demonstrate that our PromptStereo achieves state-of-the-art zero-shot generalization performance across multiple datasets, while maintaining comparable or faster inference speed. Our findings highlight prompt-guided iterative refinement as a promising direction for zero-shot stereo matching.

cs.CV↗

arXiv2026

Rethinking the Trust Region in LLM Reinforcement Learning

Reinforcement learning (RL) has become a cornerstone for fine-tuning Large Language Models (LLMs), with Proximal Policy Optimization (PPO) serving as the de facto standard algorithm. Despite its ubiquity, we argue that the core ratio clipping mechanism in PPO is structurally ill-suited for the large vocabularies inherent to LLMs. PPO constrains policy updates based on the probability ratio of sampled tokens, which serves as a noisy single-sample Monte Carlo estimate of the true policy divergence. This creates a sub-optimal learning dynamic: updates to low-probability tokens are aggressively over-penalized, while potentially catastrophic shifts in high-probability tokens are under-constrained, leading to training inefficiency and instability. To address this, we propose Divergence Proximal Policy Optimization (DPPO), which substitutes heuristic clipping with a more principled constraint based on a direct estimate of policy divergence (e.g., Total Variation or KL). To avoid huge memory footprint, we introduce the efficient Binary and Top-K approximations to capture the essential divergence with negligible overhead. Extensive empirical evaluations demonstrate that DPPO achieves superior training stability and efficiency compared to existing methods, offering a more robust foundation for RL-based LLM fine-tuning. Our code is available at https://github.com/sail-sg/Stable-RL.

cs.LG↗

arXiv2026

Revisiting Parameter Server in LLM Post-Training

Modern data parallel (DP) training favors collective communication over parameter servers (PS) for its simplicity and efficiency under balanced workloads. However, the balanced workload assumption no longer holds in large language model (LLM) post-training due to the high variance in sequence lengths. Under imbalanced workloads, collective communication creates synchronization barriers, leading to under-utilization of devices with smaller workloads. This change in training dynamics calls for a revisit of the PS paradigm for its robustness to such imbalance. We propose \textbf{On-Demand Communication (ODC)}, which adapts PS into Fully Sharded Data Parallel (FSDP) by replacing collective all-gather and reduce-scatter with direct point-to-point communication. Compared to FSDP, ODC reduces the synchronization barrier from once per layer to once per minibatch and decouples the workload on each device so that faster workers are not stalled. It also enables simpler and more effective load balancing at the minibatch level. Across diverse LLM post-training tasks, ODC consistently improves device utilization and training throughput, achieving up to a 36\% speedup over standard FSDP. These results demonstrate that ODC is a superior fit for the prevalent imbalanced workloads in LLM post-training. Our implementation of ODC and integration with FSDP is open-sourced at https://github.com/sail-sg/odc.

cs.DC↗

arXiv2025

AI4X Roadmap: Artificial Intelligence for the advancement of scientific pursuit and its future directions

Artificial intelligence and machine learning are reshaping how we approach scientific discovery, not by replacing established methods but by extending what researchers can probe, predict, and design. In this roadmap we provide a forward-looking view of AI-enabled science across biology, chemistry, climate science, mathematics, materials science, physics, self-driving laboratories and unconventional computing. Several shared themes emerge: the need for diverse and trustworthy data, transferable electronic-structure and interatomic models, AI systems integrated into end-to-end scientific workflows that connect simulations to experiments and generative systems grounded in synthesisability rather than purely idealised phases. Across domains, we highlight how large foundation models, active learning and self-driving laboratories can close loops between prediction and validation while maintaining reproducibility and physical interpretability. Taken together, these perspectives outline where AI-enabled science stands today, identify bottlenecks in data, methods and infrastructure, and chart concrete directions for building AI systems that are not only more powerful but also more transparent and capable of accelerating discovery in complex real-world environments.

physics.soc-ph↗

arXiv2025

EchoVLA: Robotic Vision-Language-Action Model with Synergistic Declarative Memory for Mobile Manipulation

Recent progress in Vision-Language-Action (VLA) models has enabled embodied agents to interpret multimodal instructions and perform complex tasks. However, existing VLAs are mostly confined to short-horizon, table-top manipulation, lacking the memory and reasoning capability required for mobile manipulation, where agents must coordinate navigation and manipulation under changing spatial contexts. In this work, we present EchoVLA, a memory-aware VLA model for mobile manipulation. EchoVLA incorporates a synergistic declarative memory inspired by the human brain, consisting of a scene memory that maintains a collection of spatial-semantic maps and an episodic memory that stores task-level experiences with multimodal contextual features. The two memories are individually stored, updated, and retrieved based on current observations, task history, and instructions, and their retrieved representations are fused via coarse- and fine-grained attention to guide base-arm diffusion policies. To support large-scale training, we further introduce MoMani, an automated benchmark that generates expert-level trajectories through multimodal large language model (MLLM)-guided planning and feedback-driven refinement, supplemented with real-robot demonstrations. Comprehensive simulated and real-world results demonstrate that EchoVLA substantially improves overall performance, e.g., it achieves the highest success rates of 0.52 on manipulation/navigation tasks and 0.31 on mobile manipulation tasks in simulation, exceeding the strong baseline $\pi_{0.5}$ by +0.20 and +0.11, respectively.

cs.RO↗

arXiv2025

Defeating the Training-Inference Mismatch via FP16

Reinforcement learning (RL) fine-tuning of large language models (LLMs) often suffers from instability due to the numerical mismatch between the training and inference policies. While prior work has attempted to mitigate this issue through algorithmic corrections or engineering alignments, we show that its root cause lies in the floating point precision itself. The widely adopted BF16, despite its large dynamic range, introduces large rounding errors that breaks the consistency between training and inference. In this work, we demonstrate that simply reverting to \textbf{FP16} effectively eliminates this mismatch. The change is simple, fully supported by modern frameworks with only a few lines of code change, and requires no modification to the model architecture or learning algorithm. Our results suggest that using FP16 uniformly yields more stable optimization, faster convergence, and stronger performance across diverse tasks, algorithms and frameworks. We hope these findings motivate a broader reconsideration of precision trade-offs in RL fine-tuning.

cs.LG↗

arXiv2025

Nonparametric Data Attribution for Diffusion Models

Data attribution for generative models seeks to quantify the influence of individual training examples on model outputs. Existing methods for diffusion models typically require access to model gradients or retraining, limiting their applicability in proprietary or large-scale settings. We propose a nonparametric attribution method that operates entirely on data, measuring influence via patch-level similarity between generated and training images. Our approach is grounded in the analytical form of the optimal score function and naturally extends to multiscale representations, while remaining computationally efficient through convolution-based acceleration. In addition to producing spatially interpretable attributions, our framework uncovers patterns that reflect intrinsic relationships between training data and outputs, independent of any specific model. Experiments demonstrate that our method achieves strong attribution performance, closely matching gradient-based approaches and substantially outperforming existing nonparametric baselines. Code is available at https://github.com/sail-sg/NDA.

cs.LG↗

arXiv2025

GEM: A Gym for Agentic LLMs

The training paradigm for large language models (LLMs) is moving from static datasets to experience-based learning, where agents acquire skills via interacting with complex environments. To facilitate this transition we introduce GEM (General Experience Maker), an open-source environment simulator designed for the age of LLMs. Analogous to OpenAI-Gym for traditional reinforcement learning (RL), GEM provides a standardized framework for the environment-agent interface, including asynchronous vectorized execution for high throughput, and flexible wrappers for easy extensibility. GEM also features a diverse suite of environments, robust integrated tools, and single-file example scripts demonstrating using GEM with five popular RL training frameworks. Along with this, we also provide a set of baselines across 24 environments using REINFORCE with Return Batch Normalization (ReBN), which -- unlike GRPO -- is compatible with the full RL setting of dense per-turn rewards and offers better credit assignment. We further conduct apple-to-apple benchmarking of PPO, GRPO and REINFORCE in both single- and multi-turn settings using GEM to shed light on the algorithmic designs. Lastly, GEM also functions as a convenient evaluation toolkit besides a training environment. We hope this framework can help accelerate future agentic LLM research.

cs.LG↗

arXiv2025

DEPTHOR++: Robust Depth Enhancement from a Real-World Lightweight dToF and RGB Guidance

Depth enhancement, which converts raw dToF signals into dense depth maps using RGB guidance, is crucial for improving depth perception in high-precision tasks such as 3D reconstruction and SLAM. However, existing methods often assume ideal dToF inputs and perfect dToF-RGB alignment, overlooking calibration errors and anomalies, thus limiting real-world applicability. This work systematically analyzes the noise characteristics of real-world lightweight dToF sensors and proposes a practical and novel depth completion framework, DEPTHOR++, which enhances robustness to noisy dToF inputs from three key aspects. First, we introduce a simulation method based on synthetic datasets to generate realistic training samples for robust model training. Second, we propose a learnable-parameter-free anomaly detection mechanism to identify and remove erroneous dToF measurements, preventing misleading propagation during completion. Third, we design a depth completion network tailored to noisy dToF inputs, which integrates RGB images and pre-trained monocular depth estimation priors to improve depth recovery in challenging regions. On the ZJU-L5 dataset and real-world samples, our training strategy significantly boosts existing depth completion models, with our model achieving state-of-the-art performance, improving RMSE and Rel by 22% and 11% on average. On the Mirror3D-NYU dataset, by incorporating the anomaly detection method, our model improves upon the previous SOTA by 37% in mirror regions. On the Hammer dataset, using simulated low-cost dToF data from RealSense L515, our method surpasses the L515 measurements with an average gain of 22%, demonstrating its potential to enable low-cost sensors to outperform higher-end devices. Qualitative results across diverse real-world datasets further validate the effectiveness and generalizability of our approach.

cs.CV↗

arXiv2025

Variational Reasoning for Language Models

We introduce a variational reasoning framework for language models that treats thinking traces as latent variables and optimizes them through variational inference. Starting from the evidence lower bound (ELBO), we extend it to a multi-trace objective for tighter bounds and propose a forward-KL formulation that stabilizes the training of the variational posterior. We further show that rejection sampling finetuning and binary-reward RL, including GRPO, can be interpreted as local forward-KL objectives, where an implicit weighting by model accuracy naturally arises from the derivation and reveals a previously unnoticed bias toward easier questions. We empirically validate our method on the Qwen 2.5 and Qwen 3 model families across a wide range of reasoning tasks. Overall, our work provides a principled probabilistic perspective that unifies variational inference with RL-style methods and yields stable objectives for improving the reasoning ability of language models. Our code is available at https://github.com/sail-sg/variational-reasoning.

cs.CL↗

arXiv2025

Language Models Can Learn from Verbal Feedback Without Scalar Rewards

LLMs are often trained with RL from human or AI feedback, yet such methods typically compress nuanced feedback into scalar rewards, discarding much of their richness and inducing scale imbalance. We propose treating verbal feedback as a conditioning signal. Inspired by language priors in text-to-image generation, which enable novel outputs from unseen prompts, we introduce the feedback-conditional policy (FCP). FCP learns directly from response-feedback pairs, approximating the feedback-conditional posterior through maximum likelihood training on offline data. We further develop an online bootstrapping stage where the policy generates under positive conditions and receives fresh feedback to refine itself. This reframes feedback-driven learning as conditional generation rather than reward optimization, offering a more expressive way for LLMs to directly learn from verbal feedback. Our code is available at https://github.com/sail-sg/feedback-conditional-policy.

cs.CL↗

arXiv2025

Continual Reinforcement Learning by Planning with Online World Models

Continual reinforcement learning (CRL) refers to a naturalistic setting where an agent needs to endlessly evolve, by trial and error, to solve multiple tasks that are presented sequentially. One of the largest obstacles to CRL is that the agent may forget how to solve previous tasks when learning a new task, known as catastrophic forgetting. In this paper, we propose to address this challenge by planning with online world models. Specifically, we learn a Follow-The-Leader shallow model online to capture the world dynamics, in which we plan using model predictive control to solve a set of tasks specified by any reward functions. The online world model is immune to forgetting by construction with a proven regret bound of $\mathcal{O}(\sqrt{K^2D\log(T)})$ under mild assumptions. The planner searches actions solely based on the latest online model, thus forming a FTL Online Agent (OA) that updates incrementally. To assess OA, we further design Continual Bench, a dedicated environment for CRL, and compare with several strong baselines under the same model-planning algorithmic framework. The empirical results show that OA learns continuously to solve new tasks while not forgetting old skills, outperforming agents built on deep world models with various continual learning techniques.

cs.LG↗

arXiv2025

SPIRAL: Self-Play on Zero-Sum Games Incentivizes Reasoning via Multi-Agent Multi-Turn Reinforcement Learning

Recent advances in reinforcement learning have shown that language models can develop sophisticated reasoning through training on tasks with verifiable rewards, but these approaches depend on human-curated problem-answer pairs and domain-specific reward engineering. We introduce SPIRAL, a self-play framework where models learn by playing multi-turn, zero-sum games against continuously improving versions of themselves, generating an automatic curriculum of stronger opponents, and eliminating the need for human supervision. To enable this self-play training at scale, we implement a fully online, multi-turn, multi-agent reinforcement learning system for LLMs and propose role-conditioned advantage estimation (RAE) to stabilize multi-agent training. SPIRAL produces reasoning capabilities that transfer broadly, improving performance by up to 10% across a suite of 8 reasoning benchmarks on 4 different models spanning Qwen and Llama model families, outperforming supervised fine-tuning on 25,000 expert game trajectories. Multi-game training (TicTacToe, Kuhn Poker, Simple Negotiation) yields the strongest results, with improvements observed across both base and instruction-tuned models. Analysis of chain-of-thought traces reveals that games develop distinct cognitive patterns that transfer to improve reasoning performance, with different games developing complementary strengths. Even models which have already been trained on reasoning tasks using RLVR, like DeepSeek-R1-Distill-Qwen-7B, still benefit from our approach. These results demonstrate that zero-sum games naturally develop transferable reasoning capabilities across diverse model architectures and training stages, highlighting a promising direction for autonomous reasoning development. Our code can be found in https://github.com/spiral-rl/spiral.

cs.AI↗

arXiv2025

PhyBlock: A Progressive Benchmark for Physical Understanding and Planning via 3D Block Assembly

While vision-language models (VLMs) have demonstrated promising capabilities in reasoning and planning for embodied agents, their ability to comprehend physical phenomena, particularly within structured 3D environments, remains severely limited. To close this gap, we introduce PhyBlock, a progressive benchmark designed to assess VLMs on physical understanding and planning through robotic 3D block assembly tasks. PhyBlock integrates a novel four-level cognitive hierarchy assembly task alongside targeted Visual Question Answering (VQA) samples, collectively aimed at evaluating progressive spatial reasoning and fundamental physical comprehension, including object properties, spatial relationships, and holistic scene understanding. PhyBlock includes 2600 block tasks (400 assembly tasks, 2200 VQA tasks) and evaluates models across three key dimensions: partial completion, failure diagnosis, and planning robustness. We benchmark 21 state-of-the-art VLMs, highlighting their strengths and limitations in physically grounded, multi-step planning. Our empirical findings indicate that the performance of VLMs exhibits pronounced limitations in high-level planning and reasoning capabilities, leading to a notable decline in performance for the growing complexity of the tasks. Error analysis reveals persistent difficulties in spatial orientation and dependency reasoning. Surprisingly, chain-of-thought prompting offers minimal improvements, suggesting spatial tasks heavily rely on intuitive model comprehension. We position PhyBlock as a unified testbed to advance embodied reasoning, bridging vision-language understanding and real-world physical problem-solving.

cs.RO↗