arXiv subjects

Siqi Zhu

Publications and source records attributed to Siqi Zhu.

17 recordsLinked to original sources

arXiv2026

Embodied Scene Rearrangement Planning

This paper introduces Embodied Scene Rearrangement Planning (ESRP), a novel task requiring embodied agents to rearrange furniture in 3D scenes to match a target configuration using only egocentric observations and a top-down target layout. Unlike prior rearrangement tasks, ESRP precludes global state access and introduces mutual object occlusions, reflecting the practical constraints of real-world robotic deployment. These factors make aligning partial egocentric observations with the global target layout particularly challenging for long-horizon planning. To facilitate research, we present ESRP-Bench, a comprehensive benchmark built on OmniGibson featuring over 5,400 scene pairs and 8,200 objects. We define three multi-level metrics to evaluate rearrangement quality and provide four baselines: a hierarchical task-and-motion planning method, a vision-language-model-based method, and two learning-based approaches (IL and RL). Experimental results demonstrate that current methods struggle to complete the task efficiently, highlighting ESRP as a challenging frontier for embodied agents in scene understanding and long-horizon task planning. This work serves as a stepping stone toward deploying intelligent agents in real-world scenarios. Project page: https://pie-lab.cn/ESRP/.

cs.RO↗

arXiv2026

Mesh-Attention: A New Communication-Efficient Distributed Attention with Improved Data Locality

Distributed attention is essential for scaling large language models (LLMs) to long contexts, yet existing methods either have limited parallelism or incur high communication costs. Ulysses uses efficient all-to-all communication but cannot scale beyond the number of attention heads, whereas Ring-Attention removes this limit at the cost of high per-GPU communication. In our analytical Q-KV interaction matrix model, Ring-Attention assigns each GPU an entire row, preserving Q locality while sacrificing KV locality. Each GPU therefore receives nearly all KV partitions, and its communication grows linearly with sequence length. We present Mesh-Attention, which instead assigns each GPU a 2D tile to balance Q and KV locality, so that it collects only subsets of Q and KV partitions. This gives Mesh-Attention asymptotically lower communication complexity than Ring-Attention without limiting parallelism. Mesh-Attention further uses KV Partition Rotation (KVR), greedy scheduling, and topology-aware GPU mapping to balance traffic, overlap communication with computation, and reduce traffic over low-bandwidth links. Across experiments on up to 256 GPUs and 1M-token sequences, Mesh-Attention achieves average speedups of 17.00x, 2.62x and 2.94x (up to 32.27x, 4.36x and 4.59x) over Ring-Attention, USP (Ulysses degree 8) and StarTrail, respectively. It maintains this performance advantage as GPU count and sequence length increase while substantially reducing communication overhead at scale.

cs.DC↗

arXiv2026

A Survey of Agent Memory in the Second Half: Towards Self-Evolving and Long-Horizon Agents

Research in artificial intelligence is shifting from model innovations and benchmark scores towards problem definition and rigorous real-world evaluation. As the field enters the "second half," the central challenge becomes real utility in long-horizon, dynamic, and user-dependent settings such as agentic coding, deep research, and computer use, where LLM-based agents face context explosion beyond fixed context windows and must continuously accumulate, manage, and selectively reuse information across extended interactions. Memory, with hundreds of papers released in 2025, therefore emerges as the critical solution to fill this utility gap. Beyond passive storage, memory is increasingly the substrate through which agents self-evolve: short-term memory gates which experiences are perceived and abstracted during execution, while long-term memory consolidates them into reusable knowledge and skills, forming the loop through which agents improve from their own experience. In this survey, we provide a unified view of foundation agent memory along three dimensions: memory substrate (internal parametric state and external retrieval-augmented stores), cognitive mechanism (sensory, working, episodic, semantic, and procedural), and memory subject (user-centric personalization and agent-centric experience). We then analyze how memory is operated under single- and multi-agent topologies and highlight learning policies over memory operations, showing how memory management itself is becoming a trainable capability spanning reinforcement-learned context curation, experience consolidation at decision time, and the emerging ecosystem of portable, shareable agent skills. Finally, we review evaluation benchmarks and metrics for memory utility, and outline open challenges and future directions.

cs.CL↗

arXiv2026

OpenTinker: Separating Concerns in Agentic Reinforcement Learning

We introduce \textsc{OpenTinker}, an open infrastructure for training large language model (LLM) agents with many LoRA-backed policies over shared execution resources. Modern agent workloads mix supervised fine-tuning (SFT), online reinforcement learning (RL), rollout generation, validation, and multi-turn environment interaction. In such workloads, LoRA adapters are not static inference artifacts: they are frequently updated policy states whose optimizer state, rollout snapshot, and training data attribution must remain consistent. \textsc{OpenTinker} centers the runtime around this policy lifecycle. Users define environments, agents, and learning objectives, while the system manages training clients, rollout samplers, checkpoint handles, and policy-version refresh. The same data path supports SFT and RL by converting trajectories into token sequences with explicit masks: context and environment observations condition the model, while generated action tokens carry supervised weights or RL advantages. This design enables multi-LoRA SFT/RL training in which many users, tasks, or agents can share a base model while keeping adapter updates, checkpoints, and rollout snapshots isolated. We describe the system architecture, the adapter lifecycle, the service-backed snapshot handoff used by the current implementation, the backend contract for mixed-adapter rollout kernels, and the training scheduler that isolates adapter-local gradients and optimizer state. Representative validation tasks exercise single-turn, multi-turn, LoRA, and multi-agent agentic training.

cs.AI↗

arXiv2026

Probing the Knowledge Boundary: An Interactive Agentic Framework for Deep Knowledge Extraction

Large Language Models (LLMs) can be seen as compressed knowledge bases, but it remains unclear what knowledge they truly contain and how far their knowledge boundary extends. Existing benchmarks are mostly static and provide limited support for systematic knowledge probing. In this paper, we propose an interactive agentic framework to systematically extract and quantify the knowledge of LLMs. Our method includes four adaptive exploration policies to probe knowledge at different granularity. To ensure the quality of extracted knowledge, we introduce a three-stage knowledge processing pipeline that combines vector-based filtering to remove strict duplicates, LLM-based adjudication to resolve ambiguous semantic overlap, and domain relevance auditing to retain valid knowledge units. Through extensive experiments, we find that Recursive Taxonomy is the most effective exploration strategy. We also observe a clear knowledge scaling law, where larger models consistently recover more knowledge. In addition, we identify a Pass@1 versus Pass@k trade-off: domain-specialized models achieve higher initial accuracy but experience rapid degradation, while general-purpose models maintain stable performance over extended extraction. Finally, our results show that differences in training data composition lead to distinct and measurable knowledge profiles across model families, reflecting how pretraining shapes each model's parametric knowledge.

cs.LG↗

arXiv2026

The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes

On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offering dense token-level supervision on trajectories sampled from the model's own policy. However, existing results on their effectiveness remain mixed: while OP(S)D has shown promise in system prompt and knowledge internalization, recent studies also report instability and degradation. In this work, we present a comprehensive empirical study of when OPD and OPSD work, when they fail, and why. We find that OPD on mathematical reasoning is highly sensitive to teacher choice and loss formulation, whereas OPSD fails in our tested settings due to test-time absence of instance-specific privileged information (PI). In contrast, OPSD is effective when PI represents a shared latent rule, such as a system prompt or alignment preference. We identify three failure mechanisms: (1) distribution mismatch between teacher and student caused by conditioning on student-generated prefixes, (2) optimization instability from biased TopK reverse-KL gradients, and (3) an OPSD-specific limitation where the student learns a PI-free policy that aggregates PI-conditioned teachers, which is insufficient when PI is instance-specific. We further show that stop-gradient TopK objectives, RLVR-adapted teachers, and SFT-stabilized students mitigate these failures.

cs.AI↗

arXiv2026

Who Prices Cognitive Labor in the Age of Agents? Compute-Anchored Wages

A natural intuition about the economics of AI agents is that, because agents can be replicated at very low marginal cost, agent labor may be supplied highly elastically, placing downward pressure on cognitive-labor wages when it closely substitutes for human labor. We argue this framing is wrong in mechanism but partially correct in conclusion, and that the correction matters for both theory and policy. \textbf{Agents are not labor; they are a production technology that converts compute capital $K_c$ into effective units of cognitive labor $L_A$.} Once this is recognized, the elastic-supply margin that anchors the equilibrium wage migrates from the labor market to the compute capital market. Building on the classic factor-pricing framework \citep{mankiw2020}, we derive a \emph{Compute-Anchored Wage} (CAW) bound stating that, on tasks where human and agent-produced cognitive labor are substitutes, the competitive human wage is bounded above by $λ\cdot k \cdot r_c$, where $r_c$ is the rental rate of compute capital, $k$ is the compute intensity of one effective agent-produced cognitive labor unit, and $λ$ is the relative human-to-agent productivity. We generalize the result through constant elasticity of substitution (CES) aggregation, separate substitutable from complementary tasks, and discuss factor-share consequences. The conclusion is concise: \emph{the price-setter for cognitive labor is no longer the labor market.}

cs.AI↗

arXiv2026

Agentic AI Systems Should Be Designed as Marginal Token Allocators

This position paper argues that agentic AI systems should be designed and evaluated as \emph{marginal token allocation economies} rather than as text generators priced by the unit. We follow a single request -- a developer asking a coding agent to fix a failing test -- through four economic layers that today are designed in isolation: a router that decides which model answers, an agent that decides whether to plan, act, verify, or defer, a serving stack that decides how to produce each token, and a training pipeline that decides whether the trace is worth learning from. We show that all four layers are solving the \emph{same} first-order condition -- marginal benefit equals marginal cost plus latency cost plus risk cost -- with different index sets and different prices. The framing is deliberately minimal: we do not propose a complete theory of AI economics. But adopting marginal token allocation as the shared accounting object explains why systems that locally minimize tokens globally misallocate them, predicts a small set of recurring failure modes (over-routing, over-delegation, under-verification, serving congestion, stale rollouts, cache misuse), and points to a concrete research agenda in token-aware evaluation, autonomy pricing, congestion-priced serving, and risk-adjusted RL budgeting.

cs.AI↗

arXiv2026

SRT: Accelerating Reinforcement Learning via Speculative Rollout with Tree-Structured Cache

We present Speculative Rollout with Tree-Structured Cache (SRT), a simple, model-free approach to accelerate on-policy reinforcement learning (RL) for language models without sacrificing distributional correctness. SRT exploits the empirical similarity of rollouts for the same prompt across training steps by storing previously generated continuations in a per-prompt tree-structured cache. During generation, the current policy uses this tree as the draft model for performing speculative decoding. To keep the cache fresh and improve draft model quality, SRT updates trees online from ongoing rollouts and proactively performs run-ahead generation during idle GPU bubbles. Integrated into standard RL pipelines (\textit{e.g.}, PPO, GRPO and DAPO) and multi-turn settings, SRT consistently reduces generation and step latency and lowers per-token inference cost, achieving up to 2.08x wall-clock time speedup during rollout.

cs.LG↗

arXiv2026

Federated Learning and Class Imbalances

Federated Learning (FL) enables collaborative model training across decentralized devices while preserving data privacy. However, real-world FL deployments face critical challenges such as data imbalances, including label noise and non-IID distributions. RHFL+, a state-of-the-art method, was proposed to address these challenges in settings with heterogeneous client models. This work investigates the robustness of RHFL+ under class imbalances through three key contributions: (1) reproduction of RHFL+ along with all benchmark algorithms under a unified evaluation framework; (2) extension of RHFL+ to real-world medical imaging datasets, including CBIS-DDSM, BreastMNIST and BHI; (3) a novel implementation using NVFlare, NVIDIA's production-level federated learning framework, enabling a modular, scalable and deployment-ready codebase. To validate effectiveness, extensive ablation studies, algorithmic comparisons under various noise conditions and scalability experiments across increasing numbers of clients are conducted.

cs.LG↗

arXiv2025

Human Motion Estimation with Everyday Wearables

While on-body device-based human motion estimation is crucial for applications such as XR interaction, existing methods often suffer from poor wearability, expensive hardware, and cumbersome calibration, which hinder their adoption in daily life. To address these challenges, we present EveryWear, a lightweight and practical human motion capture approach based entirely on everyday wearables: a smartphone, smartwatch, earbuds, and smart glasses equipped with one forward-facing and two downward-facing cameras, requiring no explicit calibration before use. We introduce Ego-Elec, a 9-hour real-world dataset covering 56 daily activities across 17 diverse indoor and outdoor environments, with ground-truth 3D annotations provided by the motion capture (MoCap), to facilitate robust research and benchmarking in this direction. Our approach employs a multimodal teacher-student framework that integrates visual cues from egocentric cameras with inertial signals from consumer devices. By training directly on real-world data rather than synthetic data, our model effectively eliminates the sim-to-real gap that constrains prior work. Experiments demonstrate that our method outperforms baseline models, validating its effectiveness for practical full-body motion estimation.

cs.CV↗

arXiv2025

GTAlign: Game-Theoretic Alignment of LLM Assistants for Social Welfare

Large Language Models (LLMs) have achieved remarkable progress in reasoning, yet sometimes produce responses that are suboptimal for users in tasks such as writing, information seeking, or providing practical guidance. Conventional alignment practices typically assume that maximizing model reward also maximizes user welfare, but this assumption frequently fails in practice: models may over-clarify or generate overly verbose reasoning when users prefer concise answers. Such behaviors resemble the prisoner's dilemma, where individually rational choices lead to socially suboptimal outcomes. The fundamental challenge is the lack of a principled decision making mechanism that mutually benefits both the LLM and the user. We propose Game-Theoretic Alignment (GTAlign), an alignment framework that integrates game-theoretic decision making into both reasoning and training. During reasoning, the model explicitly treats user-LLM interaction as a strategic game: it constructs payoff matrices within its reasoning chain to estimate welfare for both itself and the user, and then selects actions that are mutually beneficial. During training, we introduce a social welfare reward that reinforces cooperative responses, aligning model behavior with socially efficient outcomes. In addition, we introduce an inference technique that leverages game-theoretic reasoning to dynamically adapt LLM's response when pricing policies of LLM service change. Extensive experiments demonstrate that GTAlign substantially improves reasoning efficiency, answer quality, and social welfare compared to baselines across diverse tasks. The code is available at https://github.com/ulab-uiuc/GTAlign .

cs.AI↗

arXiv2025

Multi-Agent Evolve: LLM Self-Improve through Co-evolution

Reinforcement Learning (RL) has demonstrated significant potential in enhancing the reasoning capabilities of large language models (LLMs). However, the success of RL for LLMs heavily relies on human-curated datasets and verifiable rewards, which limit their scalability and generality. Recent Self-Play RL methods, inspired by the success of the paradigm in games and Go, aim to enhance LLM reasoning capabilities without human-annotated data. However, their methods primarily depend on a grounded environment for feedback (e.g., a Python interpreter or a game engine); extending them to general domains remains challenging. To address these challenges, we propose Multi-Agent Evolve (MAE), a framework that enables LLMs to self-evolve in solving diverse tasks, including mathematics, reasoning, and general knowledge Q&A. The core design of MAE is based on a triplet of interacting agents (Proposer, Solver, Judge) that are instantiated from a single LLM, and applies reinforcement learning to optimize their behaviors. The Proposer generates questions, the Solver attempts solutions, and the Judge evaluates both while co-evolving. Experiments on Qwen2.5-3B-Instruct demonstrate that MAE achieves an average improvement of 4.54% on multiple benchmarks. These results highlight MAE as a scalable, data-efficient method for enhancing the general reasoning abilities of LLMs with minimal reliance on human-curated supervision.

cs.AI↗

arXiv2025

Efficiently Scaling LLM Reasoning with Certaindex

Test-time reasoning algorithms such as chain-of-thought, self-consistency, and MCTS enhance LLM problem-solving but can wastefully generate many tokens without improving accuracy. At the same time, we observe that these algorithms exhibit answer stabilization: their intermediate solutions often cease to change after a certain point, and further investment of compute does not change their final answer. To quantify this phenomenon, we introduce Certaindex, an algorithm-agnostic metric measuring this evolving stability, signaling when further computation is unlikely to alter the final result. Certaindex is lightweight, can accelerate reasoning program inference via early exit, and further enables dynamic token allocation, gang scheduling, and many opportunities when integrated with real-world LLM serving systems. To quantify real-world benefits, we built Certaindex as a scheduler into Dynasor, our reasoning-aware LLM serving system, and demonstrate up to 50% compute savings and 3.3x higher throughput in real workloads with no accuracy drop. Our code is available at https://github.com/hao-ai-lab/Dynasor.git

cs.LG↗

arXiv2024

Efficient LLM Scheduling by Learning to Rank

In Large Language Model (LLM) inference, the output length of an LLM request is typically regarded as not known a priori. Consequently, most LLM serving systems employ a simple First-come-first-serve (FCFS) scheduling strategy, leading to Head-Of-Line (HOL) blocking and reduced throughput and service quality. In this paper, we reexamine this assumption -- we show that, although predicting the exact generation length of each request is infeasible, it is possible to predict the relative ranks of output lengths in a batch of requests, using learning to rank. The ranking information offers valuable guidance for scheduling requests. Building on this insight, we develop a novel scheduler for LLM inference and serving that can approximate the shortest-job-first (SJF) schedule better than existing approaches. We integrate this scheduler with the state-of-the-art LLM serving system and show significant performance improvement in several important applications: 2.8x lower latency in chatbot serving and 6.5x higher throughput in synthetic data generation. Our code is available at https://github.com/hao-ai-lab/vllm-ltr.git

cs.LG↗

arXiv2024

LongWriter: Unleashing 10,000+ Word Generation from Long Context LLMs

Current long context large language models (LLMs) can process inputs up to 100,000 tokens, yet struggle to generate outputs exceeding even a modest length of 2,000 words. Through controlled experiments, we find that the model's effective generation length is inherently bounded by the sample it has seen during supervised fine-tuning (SFT). In other words, their output limitation is due to the scarcity of long-output examples in existing SFT datasets. To address this, we introduce AgentWrite, an agent-based pipeline that decomposes ultra-long generation tasks into subtasks, enabling off-the-shelf LLMs to generate coherent outputs exceeding 20,000 words. Leveraging AgentWrite, we construct LongWriter-6k, a dataset containing 6,000 SFT data with output lengths ranging from 2k to 32k words. By incorporating this dataset into model training, we successfully scale the output length of existing models to over 10,000 words while maintaining output quality. We also develop LongBench-Write, a comprehensive benchmark for evaluating ultra-long generation capabilities. Our 9B parameter model, further improved through DPO, achieves state-of-the-art performance on this benchmark, surpassing even much larger proprietary models. In general, our work demonstrates that existing long context LLM already possesses the potential for a larger output window--all you need is data with extended output during model alignment to unlock this capability. Our code & models are at: https://github.com/THUDM/LongWriter.

cs.CL↗

arXiv2019

Optical Superoscillatory Poisson-Arago Spots

Optical diffraction limit has been a long-term scientific issue since Ernst Abbe first introduced the concept in 1873. It is a constraint on the smallest light spot that can be achieved. Substantial effort has been invested in the past decade to beat this limit by exploiting evanescent waves. But this method encounters serious near-field limitations. A more promising route to breaking the constraint is to explore optical superoscillation in the far field with engineered metamaterials. However, these particular structures involve with very complicated optimization-based design that requires precisely tailoring the interference of propagating waves with low spatial frequency. To overcome these limitations, here we explore a new approach based on the two-hundred-year-old discovery: Possion-Arago spots. We show for the first time that by using a single disc, constructive interference of propagating waves with high-spatial-frequency wavevectors can be realized, generating a diffraction-unlimited localized Possion-Arago spot with achievable size down to $λ/$20. Actually, such an element permits creation of an ultra-long nearly nondiffracting superoscillatory needle with appreciable field of view. This easy-to-fabrication element provides a promising route to overcome the diffraction limit, thus might open new avenues to exploit various applications in different fields.

physics.optics↗