Search arXiv⌕ Search

arXiv subjects

He Li

Publications and source records attributed to He Li.

At least 19 recordsLinked to original sources
arXiv2026

StateLoop: Persistent State for Looped Transformers

Looped Transformers offer a parameter-efficient approach to language modeling by repeatedly applying a shared backbone. Vanilla looped Transformers use each backbone output directly as the input to the next iteration. These outputs may contain information useful several iterations later, but retaining it across the loop depends entirely on the single hidden state. We introduce StateLoop, which maintains one additional persistent state per token to retain information across iterations. At each iteration, a learned gate controls how much of the previous state is retained and how much information from the backbone output is incorporated. The updated state is projected back and added to the backbone output for subsequent iterations and final prediction. This design reduces Pile perplexity and improves average accuracy across eight zero-shot tasks over Vanilla Looped. Its perplexity advantage over Vanilla Looped grows with trained loop depth, consistent with persistent state becoming increasingly useful as recurrent computation deepens.

cs.CL↗
arXiv2026

LoopFold: Parallelizing Recurrent Depth in Looped Transformers

Looped Transformers increase computation per token by repeatedly applying a shared backbone, offering a way to improve language modeling and downstream task performance at a fixed backbone parameter count. However, each loop depends on the preceding loop, so increasing recurrent depth also increases decoding latency. Existing approaches include Parallel Loop Transformers, which reduce latency by overlapping loop computation across token positions, but yield rapidly diminishing marginal accuracy gains at larger depths. To address this challenge, we introduce LoopFold, which folds computation at multiple recurrent depths into parallel batches for the same token prefix. A lightweight predictor estimates inputs to later loops so they can start before earlier loops finish. A corrector then refines the outputs of these later loops using newly available results from earlier loops. All components are trained jointly using only the final next-token prediction loss. In experiments with 1.4B models, deeper LoopFold configurations achieve progressively higher average zero-shot accuracy across eight tasks, rising from 53.41% to 56.38%. Vanilla looped models and PLT show limited gains over the same depth range. Compared with a nine-loop vanilla model, LoopFold uses five sequential stages and additional parallel computation to reduce decoding latency by 46%.

cs.CL↗
arXiv2026

P2N: Reusing Deep Representations for Greater Effective Depth

Mathematical problem solving and autonomous agents are increasingly important applications of language models, demanding stronger reasoning capabilities. Computational depth is a key resource for such reasoning, but adding model layers increases both training and inference costs, while test-time scaling incurs substantial additional inference computation. We introduce P2N (Previous to Next), which reuses the previous token's deep representations, already available during generation, to increase effective computational depth with negligible decoding overhead. Specifically, P2N adds the previous token's deep hidden state to the current token's shallow representation, allowing subsequent layers to build on previously completed deep computation. This recurrence extends effective computational depth across tokens, while introducing only a single vector addition per decoding step. We train the recurrence using parallel Jacobi updates restricted to a middle-layer core. Experiments with Qwen-style models show that, at 1.5B parameters, P2N improves GSM8K accuracy by over 10 percentage points under both zero-shot and few-shot prompting, with nearly unchanged decoding latency. The benefits also extend to language modeling and average performance on general downstream benchmarks. on general downstream benchmarks.

cs.CL↗
arXiv2026

Dual-Slot: From Serial Loops to Parallel Slots in Language Models

Looped language models increase computational depth by repeatedly applying shared layers, offering a way to improve model quality without adding parameters. However, sequential dependencies between loop iterations increase autoregressive decoding latency. We introduce Dual-Slot, an architecture that reorganizes serial looped computation into parallel latent and prediction slots. Rather than waiting for the current token's latent computation to finish, the prediction slot combines the previous token's deep latent state with the current representation, while the latent slot produces a state for the next decoding step. This shifts the latent dependency across adjacent tokens, replacing two sequential module invocations with a single paired invocation while preserving standard autoregressive generation. To preserve token-level parallelism during training, we use a small number of Jacobi iterations to refine latent states across the sequence in parallel. Experiments on Qwen-style dense and mixture-of-experts (MoE) models show consistent improvements in language modeling and average downstream accuracy. On the 1B dense model, Dual-Slot reduces Pile validation perplexity from 7.25 to 7.00 and improves average five-shot accuracy across nine task configurations by 2.26 percentage points. After supervised fine-tuning, it improves zero-shot GSM8K accuracy by 5.83 and 3.49 percentage points for the 1B dense and A0.6B MoE models, respectively. Across the evaluated batch sizes, Dual-Slot retains 93.1-96.5% of vanilla decoding throughput for the dense model and 98.1-99.6% for the MoE model.

cs.CL↗
arXiv2026

Quantifying reticulocyte biomechanics in health and disease

Red blood cell (RBC) populations are mechanically and morphologically heterogeneous, yet how this heterogeneity governs transport, splenic retention and blood rheology in confinement remains unclear. We combine microchannel experiments with dissipative particle dynamics (DPD) simulations to determine how the morphology, deformability and hydrodynamic interaction of immature RBCs (reticulocytes) shape microconfined flow in acute and chronic mountain sickness. Reticulocyte-rich samples present the three shapes that the maturation literature assigns to successive reticulocyte stages -- multilobular, cup-shaped and near-discocytic -- which we model as R1-R3, each with a membrane shear modulus within the range reported for reticulocytes. A 5-um microchannel resolves these models only weakly, R1 taking at most ~18% longer than control discocytes to cross it, whereas a 1.5-um splenic slit, driven at its own loading, delays R1 by ~30%; both geometries order the models by shear modulus. In cell pairs, a leading cell never lets a follower pass below its own single-cell threshold, ruling out the order-of-magnitude reduction that a wake-"unjamming" picture would suggest; instead the follower queues behind the leader, and a stiffer leader, which clears the slit more slowly, delays it more. The controlling variable is the single-cell critical pressure gradient Delta P_c, which rises monotonically from control discocytes through the reticulocyte subtypes to sickled cells. Set against published rheograms, an analytical estimate attributes chronic-mountain-sickness hyperviscosity mainly to hematocrit-driven crowding rather than single-cell rheology. These results place benign acclimatization, chronic-mountain-sickness hyperviscosity and sickle-cell-trait splenic syndrome on a single mechanical axis defined by Delta P_c.

cond-mat.soft↗
arXiv2026

The typical red cell makes too little vacuole material for coalescence to build the large vacuoles of pocked cells

Red blood cells carrying intracellular vacuoles, called pocked cells, are a sensitive marker of splenic function: about 0.5% of cells in healthy people, 20 to 70% after splenectomy. Vacuoles too small to see are common, and visible ones are thought to grow when small ones merge. We tested that idea with a particle model that resolves the cell's membrane, its contents and every vacuole, and with an inventory of the vacuole volume a cell can make. In tube, shear and vascular flows, a cell's vacuoles met within a second, and, allowed to fuse, ten merged into one to five. Meeting is not what limits growth; material is. At the rate measured after splenectomy, all the vacuoles an average cell makes in its 120-day life fuse into one body 0.6 $μ$m across, the size healthy cells carry, but 16 times smaller in volume than the 1.5-$μ$m bodies of patients without a spleen. Taking vacuole size as set at formation, a population model in which vacuoles are made mainly in young cells, and far faster in a small minority that alone makes material for a large one, accounts for the published counts, sizes and age profiles and, on first-hand measurements, favours a splenic clearance that rises several-fold as cells age; letting that minority's vacuoles fuse fits nearly as well. At a model splenic slit, the cell's spare surface decided passage, and a vacuole mattered only when it led. A slit experiment would test whether brief encounters fuse vacuoles.

physics.bio-ph↗
arXiv2026

Forecasting Without Full Identification: Reliability and Computational Cost in Clot-Growth Simulation

We forecast clot growth from sparse early measurements with incomplete biochemical information. Using 1,000 simulated trajectories, we compare inverse search, which weights candidate inputs by their agreement with early observations, with a direct multilayer perceptron (MLP) predicting future clot size. Inverse search is more accurate when tissue factor alone, or tissue factor and fibrinogen, are unknown. When all five inputs are unknown, direct prediction achieves 2.16% pooled relative $L_2$ forecast error, outperforming search with 8,192 candidates using a forward MLP (3.25%) or latent neural ordinary differential equation model (3.36%). A conditional error bound separates candidate disagreement, weight on poor early fits, surrogate error, and candidate coverage. Increasing randomized quasi-Monte Carlo candidate budgets from 512 to 32,768 reduces numerical variation. A covariance score ranks additional assays by expected forecast value while accounting for measurement noise. At 10% range-scaled assay noise, incorporating assay uncertainty reduces forecast error from about 5% to below 3% for both surrogates. An independent reaction cascade verifies candidate integration against tensor quadrature, with mean normalized forecast-integration error below $10^{-5}$ at 32,768 candidates. These results demonstrate accurate clot-growth forecasting without full biochemical identification and support selecting prediction strategies and additional measurements according to accuracy, reliability, and computational cost.

cs.LG↗
arXiv2026

SpecFold: Folding Multi-Branch Redundancy for Faster Speculative Decoding in Diffusion Language Models

Diffusion large language models (DLLMs) generate text through iterative block denoising, and multi-branch speculative decoding accelerates this process by verifying a main branch together with multiple draft branches in a single forward pass. While prior DLLM acceleration methods primarily exploit temporal redundancy across denoising steps, we identify a complementary redundancy axis within each speculative verification step: multi-branch computational redundancy. During speculative verification, draft branches inherit most tokens from their parents while unmasking a small set of additional positions, causing large portions of hidden states to remain highly similar across branches. We propose SpecFold, an algorithm-system co-design that exploits this multi-branch redundancy to reduce the cost of multi-branch speculative verification. Algorithmically, SpecFold performs token-level residual gating and selectively reuses parent computation through folded attention and FFN while preserving residual hidden states. Systemically, a Triton kernel implementation translates this fine-grained reuse into end-to-end throughput gains through efficient sparse multi-branch execution. SpecFold is orthogonal to temporal caching and compatible with existing DLLM speculation strategies. Across two DLLM families, five models, and five standard benchmarks, SpecFold achieves up to 1.64x throughput over Spiffy and up to 1.99x over vanilla decoding, while maintaining comparable task performance.

cs.AI↗
arXiv2026

Self-Evolutionary Replanning for Failure-Aware Motion Planning

Motion planners tuned under nominal conditions can fail when execution conditions change. Existing differentiable adaptation methods provide efficient online updates but remain constrained by static loss functions and gradient-only parameter updates. We propose Self-Evolutionary RePlanning (SERP), a failure-aware framework that jointly adapts motion-planner parameters and loss function weights from closed-loop execution feedback. SERP provides two adaptation mechanisms. First, in-context learning with auto-differentiation (ILAD) enables rapid online adaptation by coupling local numerical refinement with history-conditioned semantic intervention. Auto-differentiation performs numerical updates, while a vision-language model (VLM) interprets recent adaptation history to jointly revise planner parameters and loss weights when local adaptation becomes ineffective. Second, the memory-augmented rollout system (MARS) extends ILAD into a deliberative and evolutionary recovery process with memory-guided candidate generation and parallel low-fidelity rollout evaluation. The resulting recovery trace is then consolidated into recovery memory to guide future adaptation. Extensive simulation and real-world experiments demonstrate improved failure recovery performance over various baselines across diverse environments.

cs.RO↗
arXiv2026

Evaluation of OpenAI o1: Opportunities and Challenges of AGI

This comprehensive study evaluates the performance of OpenAI's o1-preview large language model across a diverse array of complex reasoning tasks, spanning multiple domains, including computer science, mathematics, natural sciences, medicine, linguistics, and social sciences. Through rigorous testing, o1-preview demonstrated remarkable capabilities, often achieving human-level or superior performance in areas ranging from coding challenges to scientific reasoning and from language processing to creative problem-solving. Key findings include: -83.3% success rate in solving complex competitive programming problems, surpassing many human experts. -Superior ability in generating coherent and accurate radiology reports, outperforming other evaluated models. -100% accuracy in high school-level mathematical reasoning tasks, providing detailed step-by-step solutions. -Advanced natural language inference capabilities across general and specialized domains like medicine. -Impressive performance in chip design tasks, outperforming specialized models in areas such as EDA script generation and bug analysis. -Remarkable proficiency in anthropology and geology, demonstrating deep understanding and reasoning in these specialized fields. -Strong capabilities in quantitative investing. O1 has comprehensive financial knowledge and statistical modeling skills. -Effective performance in social media analysis, including sentiment analysis and emotion recognition. The model excelled particularly in tasks requiring intricate reasoning and knowledge integration across various fields. While some limitations were observed, including occasional errors on simpler problems and challenges with certain highly specialized concepts, the overall results indicate significant progress towards artificial general intelligence.

cs.CL↗
arXiv2026

A System Architecture for Low Latency Multiprogramming Quantum Computing

As quantum systems scale, multiprogramming quantum computing (MPQC) provides a practical way to improve device utilization and throughput. However, because quantum executables are device-dependent, non-portable across qubit regions, and highly susceptible to noise and crosstalk, current MPQC pipelines rely on expensive online compilation to co-optimize concurrently running programs. This online step dominates runtime and impedes low-latency deployments for practical, real-world workloads in the future, such as repeatedly invoked quantum neural network (QNN) services. We present FLAMENCO, a fidelity-aware multi-version compilation system that enables independent offline compilation and low-latency multiprogramming at runtime. \textbf{At the architecture level}, the system abstracts devices into compute units to reduce the search space of region allocation. \textbf{At compile time}, it generates diverse executable versions for each program---each bound to a distinct qubit region---allowing dynamic region selection at runtime and overcoming non-portability. \textbf{At runtime}, it employs a lightweight orchestrator that uses post-compilation fidelity metrics to avoid conflicts and mitigate crosstalk, supporting conflict-free co-execution without online co-optimization. Evaluations show that FLAMENCO achieves over 5$\times$ runtime speedup in post-scheduling execution while maintaining comparable execution fidelity on common-success workloads. When integrated into existing scheduler-coupled systems, it raises workload-level conflict-free orchestration ratio from 0.183 to 1.000 for HyperQ and from 0.050 to 0.400 for QOS.

cs.AR↗
arXiv2026

Population-Scalable Multi-Agent World Modeling

World models have recently achieved impressive progress in visual prediction and interactive generation, but extending them to multi-agent environments introduces a fundamental scalability challenge. Existing methods generally assume a fixed number of agents during training and inference, which ties the model to a pre-determined agent population and limits inference-time scalability. Our key insight is that cross-view consistency should arise from a shared world state whose evolution does not assume a predefined number of agents, while agent-specific observations should be generated by querying this state through a unified rendering interface. Based on this insight, we propose Khora, a scalable multi-agent world model that supports inference-time expansion to arbitrary numbers of agents without retraining. Our framework decouples world-state evolution from visual rendering and introduces a population-agnostic rendering mechanism for incorporating other agent information. This design maintains cross-view consistency through the shared world state rather than through dense interactions among observation streams inside the expensive video generator, enabling approximately linear practical scaling with the number of queried views. Qualitative experiments demonstrate that our approach generalizes to unseen numbers of agents while maintaining visual quality and multi-agent consistency. We further implement a real-time interactive system to demonstrate scalable open-world simulation.

cs.CV↗
arXiv2026

RECAP: Feedback-Driven Streaming Semantic User Profiles for Short-Video Recommendation

Language-based user profiles convert long behavioral histories into explicit semantic representations for recommendation. However, most profile generators are optimized in an open loop: they may summarize past behavior fluently, but are not directly trained to improve future recommendation. We study this problem in real-world short-video recommendation, where user behaviors continuously arrive as streams and profiles must be incrementally updated under limited capacity. This requires maintaining a consistent bounded profile state and constructing profile-targeted semantic feedback from industrial implicit behavior logs. We propose RECAP, an offline closed-loop framework for optimizing streaming structured semantic profiles with historical recommendation feedback. RECAP maintains each profile as a bounded structured memory by combining LLM-based semantic updates with deterministic lifecycle and capacity control. RECAP constructs profile-targeted semantic feedback by filtering label-consistent behavior pairs with an LLM judge and training a dual-tower evaluator whose matching score serves as a GRPO reward. Experiments on Kuaishou short-video data show that RECAP improves uAUC by 0.0084 and Recall@2000 by about 4.9% over the base generator. Further analyses confirm the benefits of feedback construction and policy optimization, and show more grounded refinement and user-level abstraction in profile updates. A seven-day online A/B test further shows a statistically significant 0.139% improvement in average application usage time per user.

cs.IR↗
arXiv2026

SpaCellAgent: A Self-Evolving LLM-Based Multi-Agent Framework for Trajectory Analysis

Spatial and Single-cell transcriptomics are transformative in deciphering cellular dynamics. As the fundamental paradigm for reconstructing cell developmental paths, trajectory inference (TI) is critical. However, existing methods require extensive manual intervention and proficiency in heterogeneous tools, posing a significant barrier to efficient TI analysis. To bridge this gap, we propose SpaCellAgent, an autonomous large language model (LLM) multi-agent framework that automates end-to-end spatiotemporal analysis and narrative generation. SpaCellAgent utilizes a multi-agent architecture for strategic workflow planning, a dynamic tool-orchestration engine for adaptive algorithm selection, and a self-evolution module that iteratively refines performance through feedback. We evaluate SpaCellAgent on six heterogeneous datasets encompassing complex temporal developmental trajectories, diverse sequencing platforms, and spatially-resolved tissue architectures. SpaCellAgent consistently demonstrates over 40\% improvement in analytical efficiency while maintaining expert-aligned performance. By converting natural language specifications into optimized analytical workflows and fully automating the pipeline, SpaCellAgent democratizes advanced spatiotemporal modeling and establishes a scalable, agent-driven paradigm for computational biology. The code and materials are available at https://github.com/LittleXH-shw/SpaCellAgent.

cs.AI↗
arXiv2026

A Geometric Analysis-Based Safety Assessment Framework for Marine Vehicle Route Decision-Making

This paper develops a Geometric Analysis-based Route Safety Assessment (GARSA) framework to enhance the safety of marine vehicles navigating in restricted waters. Utilizing line and point geometric elements to define waterway boundaries, the framework enables the construction of a dynamic width characterization function to quantify spatial safety within complex restricted navigation spaces. An iterative method is developed to calculate the dynamic width characterization function, enabling an abstracted spatial property representation of waterways. Based on this, a navigational safety index that accounts for the overall route safety as well as the spatial constraints of local segments is created, enabling the selection of the safest waterway to pass through. Case studies in Hamburg Port and hypothetical waterways show that GARSA identifies spatial safety differences among routes. For example, GARSA assigned safety assessment values of 0.6 and 135 to two intra-port transit routes of 7,258 m and 7,845 m, respectively, with the higher value indicating a safer route. Overall, GARSA provides a quantitative basis for safety-oriented route decision-making of marine vehicles in restricted waters.

eess.SY↗
arXiv2026

Depth-Attention: Cross-Layer Value Mixing for Language Models

Self-attention selects information freely across the sequence, but across depth, Transformers merely add each layer's output to the residual stream, so later layers cannot selectively reuse earlier-layer representations. Recent cross-layer methods improve this flow but operate on hidden states outside attention, adding state beyond the key-value cache at inference--a cost that becomes increasingly salient as modern LLMs compress the cache with grouped-query and multi-head latent attention. We introduce Depth-Attention, which performs this selection inside the attention module itself: before a layer attends over the sequence, its query attends over the keys of earlier layers at the same token position and mixes their values into the value that self-attention then reads. Because Depth-Attention reuses the standard attention queries, keys, and value-cache slots, storing depth-mixed values in place of the original values, it adds no parameters and introduces no persistent inference state beyond the standard key-value cache--the same cache size as a vanilla decoder and less than hidden-state-based cross-layer methods. On Qwen3-style decoders at 1.5B and 3B parameters, Depth-Attention attains the lowest perplexity and the highest average downstream accuracy, improving over the vanilla Transformer by up to 2.3 accuracy points and surpassing strong cross-layer baselines in perplexity and average accuracy, while adding under 0.01% extra arithmetic FLOPs and no additional persistent inference state. The gains hold from 360M to 3B parameters and extend to looped Transformers.

cs.CL↗
arXiv2026

MLUBench: A Benchmark for Lifelong Unlearning Evaluation in MLLMs

Multimodal large language models (MLLMs) are trained on massive multimodal data, making data unlearning increasingly important as data owners may request the removal of specific content. In practice, these requests often arrive sequentially over time, giving rise to the challenging problem of MLLM Lifelong Unlearning. However, most existing benchmarks are limited in scale and scope, failing to capture the complexities of MLLM lifelong unlearning. To fill this gap, we introduce the MLUBench, a large-scale and comprehensive benchmark featuring 127 entities across 9 classes under lifelong unlearning requests. We perform extensive experiments using MLUBench and reveal that existing unlearning methods suffer from severe, cumulative degradation. More critically, we further identify the unique challenge of this problem: unlike in unimodal models, MLLM lifelong unlearning is constrained by the need to preserve multimodal alignment. Continually unlearning from one modality could degrade the entire model. To alleviate this challenge, we propose LUMoE, an effective method. Experiments demonstrate that LUMoE significantly mitigates the degradation problem faced by baselines. The source code and the MLUBench dataset are open-sourced in https://github.com/lihe-maxsize/Lifelong_Unlearning_main.

cs.AI↗
arXiv2026

Quantifying the biophysical properties of stomatocytes in health and disease

Hereditary stomatocytosis (HS) comprises red blood cell (RBC) disorders characterized by cup-shaped erythrocytes that respond oppositely to splenectomy: curative in overhydrated HS (OHS) but potentially thrombogenic in dehydrated HS (DHS/xerocytosis). This paradox persists because RBC biomechanics is governed by partly independent parameters--shear modulus, bending rigidity, surface-to-volume ratio (S/V), and cytoplasmic viscosity--that existing assays capture only piecemeal. Here we combine dissipative particle dynamics (DPD) simulations with microfluidic imaging to construct a control discocyte and three stomatocyte models (ST-RBC1-3) at fixed membrane area and decreasing volume (109.7, 101.5, 89.8 fL), spanning the OHS-to-DHS range. Tracing this parameter set through five mechanically orthogonal assays, we find that interendothelial-slit (IES) traversal is geometry-dominated: overhydrated ST-RBC1 requires an order of magnitude higher critical pressure than healthy RBCs, whereas dehydrated ST-RBC3 passes freely. ST-RBC3 nonetheless suppresses membrane tank-treading and raises low-shear whole-blood viscosity by ~29% at physiological haematocrit, comparable to Gaucher-disease hyperviscosity. A funnel-obstacle chip amplifies these differences into a label-free centerline-offset signal predicted to separate all four RBC types (~4.5 standard deviations between extreme phenotypes). These results unite single-cell mechanics, splenic filtration, and hemorheology in one framework, resolve the splenectomy paradox, and point toward microfluidic pre-operative risk stratification in HS.

q-bio.CB↗