Searcharxiv⌕ Search

arXiv subjects

Ang Li

Publications and source records attributed to Ang Li.

At least 19 recordsLinked to original sources
arXiv2026

A Multidimensional Evaluation Framework for Artificial Intelligence-Based Ocean State Forecasts

Artificial intelligence (AI) based ocean state forecasts now approach or surpass those produced by operational ocean forecasting systems as measured by conventional pointwise metrics. However, these metrics may favor spatially smoothed forecasts and cannot determine whether forecast variables remain physically coherent. Forecast evaluation should therefore move beyond overall pointwise accuracy toward assessing forecast trustworthiness. We propose a multidimensional framework comprising overall pointwise accuracy, physical consistency, and distributional consistency. Overall pointwise accuracy summarizes forecast and reference agreement over all valid samples; physical consistency evaluates dynamical balances using model outputs alone; and distributional consistency evaluates distribution tails and multiscale spatial structures. Three AI models were deployed as representative testbeds in a unified operational environment. This comprehensive assessment shows that lower overall pointwise errors cannot preclude substantial dynamical imbalances and suppressed fine-scale variance. Conversely, higher pointwise errors may arise from spatial displacement of coherent structures rather than a poor representation of their amplitude. The results demonstrate that the evaluation of AI ocean forecasting models should be based on the multi-dimensional assessment of forecast trustworthiness, rather than solely relying on a single metric. The study provides a preliminary framework for AI model intercomparison, which is also informative for targeted model refinement and operational implementation.

physics.ao-ph↗
arXiv2026

Library of Layers (LoL): Adaptively Ordering Transformer Layers at Runtime

Computer programmers chain together the same reusable computational primitives, for example, conditional statements or loops, but in different orders to implement different algorithms. In contrast, existing language model architectures always execute their layers in the same order. We introduce Library of Layers (LoL), an architecture that dynamically orders its layers during inference to solve the problem at hand. LoL has a single hyperparameter, the size of its global library of transformer blocks, that allows it to transition seamlessly between parameter reuse and sparsity, generalizing the parameter-efficiency and FLOPs-efficiency tradeoffs of dense, depth-recurrent, and mixture of experts architectures in one unified family. We show that LoL achieves competitive performance with all three above baselines when pretrained on text data. On synthetic sudoku and maze data, we observe that the model employs distinct layer orderings to implement different solution strategies, and we can intervene on routing to force a model to apply a specific solution strategy to a problem instance. In a multilingual natural language setting, we find that LoL chooses different layer order patterns based on language and script. Similarly, we train LoL on safety data and fit probes to its router outputs, finding that these probes distinguish safe from unsafe text with high AUROC. LoL provides a path toward a future in which models dynamically compose their own computational graph, enabling more efficient and controllable reasoning.

cs.LG↗
arXiv2026

Step-dLLM: Adaptive Step-aware Sparse Attention for Efficient Diffusion LLM Inference

Diffusion large language models (dLLMs) generate tokens in parallel via iterative denoising, but their bidirectional attention prevents the KV caching used by autoregressive models and recomputes the full attention map at every step, incurring quadratic per-step cost. Existing sparse attention methods for dLLMs construct a sparse mask after an initial full-attention warm-up and reuse it for the remaining steps, while allocating the sparsity budget uniformly across heads. We argue both choices ignore key structural properties of attention in dLLMs: empirically, attention patterns are only locally consistent across steps and shift abruptly at certain transitions, and different heads concentrate their attention mass to very different extents. To this end, we design Step-dLLM, which periodically refreshes a block-sparse mask at anchor steps along the denoising trajectory and reuses it within each locally consistent interval, while employing a head-adaptive global budget that more aggressively compresses heads with concentrated attention. Backed by a fused IO-aware Triton kernel for block-sparse attention, Step-dLLM applies to pre-trained LLaDA-1.5 and Dream-7B-Instruct without retraining, achieving a 2.78x end-to-end speedup over dense FlashAttention on LLaDA-1.5 at 8K context with 256 denoising steps and 90% sparsity, while improving the accuracy-latency trade-off over prior sparse-attention methods.

cs.CL↗
arXiv2026

Splaxel: Efficient Distributed Training of 3D Gaussian Splatting for Large-scale Scene Reconstruction via Pixel-level Communication

3D Gaussian Splatting (3DGS) enables high-fidelity and real-time 3D scene reconstruction, but scaling training to large-scale scenes requires optimizing hundreds of millions of Gaussians across multiple GPUs. Existing distributed approaches either partition scenes into isolated regions, causing global inconsistency, or rely on global Gaussian-level exchanges, which lead to substantial growth in inter-GPU communication and quickly dominate iteration time. We propose Splaxel, a communication-efficient distributed 3DGS training framework based on pixel-level local rendering and global composition. Instead of synchronizing Gaussians, each GPU renders its local subset and exchanges only partial pixel values, maintaining mathematical consistency while keeping communication cost stable as the scene size increases. Splaxel further reduces pixel-level redundancy through geometric and transmittance visibility prediction and improves GPU utilization via conflict-free camera-view consolidation. Evaluated on large-scale datasets with up to 120M Gaussians, Splaxel achieves up to 7.6$\times$ speedup over the state-of-the-art distributed 3DGS framework while preserving high reconstruction quality.

cs.DC↗
arXiv2026

Early Fault-Tolerant Magic State Synthesis and Spacetime Efficient Analog Rotations with Color Code on a Superconducting Quantum Processor

With recent improvements to syndrome extraction scheduling and decoding, and the availability of complete transversal Clifford operations and lattice surgery, color code is an especially promising platform for quantum processors with local connectivity. The key ingredient for universal computation is thus generating useful magic states with limited resources. Space-Time Efficient Analog Rotations (STAR) and Fault-tolerant magic state synthesis (MSS) are two such methods for magic state generation, targeted for near-term and long-term quantum computing, respectively. In this work, we perform the first implementations of both magic state paradigms on a superconducting quantum processor. We present a STAR-type injection protocol on the color code, with favorable error rate and spacetime costs versus prior protocols on both the color and surface codes. We then implement a MSS factory by performing multi-patch magic state cultivations, interleaved with transversal Clifford operations, and lattice-surgery magic state teleportation. Under post-selection, we achieve a kickback tomography correct sign probability of $99.9859\%$ on cultivated $T_L$ states, $96\%$ tomographic fidelity on logical analog rotation states, and $91\%$ fidelity on synthesized logical magic states consuming 3 cultivated $T_L$ states, 3 transversal Hadamard operations, and real-time feed-forward $S_L$ corrections during magic state teleportation. We compare the efficacy of post-selection, Most-Likely-Error (MLE) decoding, and concatenated-matching decoding throughout, highlighting the potential of end-to-end color code computation.

quant-ph↗
arXiv2026

Epsilon-Identifiability of Causal Quantities

Identifying the effects of causes and causes of effects is vital in virtually every scientific field. Often, however, the needed probabilities may not be fully identifiable from the available data sources. This paper shows how approximate identifiability is still possible for several probabilities of causation. We term this epsilon-identifiability and demonstrate its usefulness in cases where the behavior of certain subpopulations can be restricted within sufficiently narrow bounds. In particular, we show how unidentifiable causal effects and counterfactual probabilities can be epsilon-identified when such allowances are made. Often, these allowances are easily measured and reasonably assumed. Finally, epsilon-identifiability is applied to the unit selection problem.

cs.AI↗
arXiv2026

Fully Parallelized BP Decoding for Quantum LDPC Codes Can Outperform BP-OSD

This work presents a hardware-efficient and fully parallelizable decoder for quantum LDPC codes that leverages belief propagation (BP) with a speculative post-processing strategy inspired by classical Chase decoding algorithm. By monitoring bit-level oscillation patterns during BP, our method identifies unreliable bits and generates multiple candidate vectors to selectively flip syndromes. Each modified syndrome is then decoded independently using short-depth BP, a process we refer to as BP-SF (syndrome flip). This design eliminates the need for costly Gaussian elimination used in the current BP-OSD approaches. Our implementation achieves logical error rates comparable to or better than BP-OSD while offering significantly lower latency due to its high degree of parallelism for a variety of bivariate bicycle codes. Evaluation on the $[[144,12,12]]$ bivariate bicycle code shows that the proposed decoder reduces average latency to approximately 70% of BP-OSD. When post-processing is parallelized the average latency is reduced to 55% compared to the single process implementation, with the maximum latency reaching as low as 18%. These advantages make it particularly well-suited for real-time and resource-constrained quantum error correction systems.

quant-ph↗
arXiv2026

An Attribute-Based Measure of Video Complexity

A new framework for the estimation of the complexity posed by video-question pairs to video-LLMs, Video Attribute-Based Complexity (VideoABC), is proposed. Video complexity is defined as the probability of failure of a video-LLM for a given video-question pair. VideoABC is a non-parametric complexity measure, using a reference video dataset and a pre-defined vocabulary of video attributes informative of complexity, e.g. the scene complexity or the speed of the video event relevant to the question. In a training phase, reference videos are projected into the space of these attributes, which is then quantized. The expected ABC of each quantization cell is then computed. Given a new video and its projection into the attribute space, complexity is estimated by the expected ABC of the associated quantization cell. To enable the use of VideoABC with small reference video datasets, two quantizers are combined: a k-means quantizer that enables accurate complexity estimates for samples in the distribution of the reference dataset and a universal lattice quantizer that guarantees generalization to out-of-distribution samples. A synthetic video generation procedure, inspired by target-distractor manipulations of psychophysics studies, is proposed to populate the cells of the lattice quantizer during training, enabling the computation of their expected ABCs. Experimental results show that VideoABC is effective even with very low-dimensional attribute representations, substantially outperforming approaches like 'video-LLM as judge' with much less complexity. Finally, the explainable nature of the VideoABC score, in terms of well-defined attributes, is shown to provide insights on how the attribute composition of benchmarks affects their complexity.

cs.CV↗
arXiv2026

Forward-Only Learning for Shared-Kernel Wavelet Neural Networks without Backpropagation

Backpropagation remains the dominant training paradigm, using backward gradients and local updates that can suffer from vanishing or exploding gradients in deep networks. Although the Forward-Forward algorithm avoids conventional backpropagation, it remains mainly local or layer-wise, motivating global whole-network learning using forward evaluations only. For shared-kernel wavelet neural networks, we therefore propose a CMA-ES-based global forward-only learning framework. We further introduce structure-aware parameter decomposition and channel-specific credit assignment, factorizing the search into three subspaces. On BSDS500, our method reduces test MSE from 4.1783 to 4.0130 and improves PSNR from 42.621 to 42.787~dB over Adam/BP without backward propagation. Beyond these gains, training requires only 6 training-set coverages, compared with 1500 epochs for BP. Ablations show that structural decomposition and aligned credit assignment are more important than increasing population size, demonstrating the feasibility of whole-network forward-only learning for compact reconstruction networks. These results suggest a potential route toward forward-only optimization of compact reconstruction models for resource-constrained imaging and edge or near-sensor systems.

cs.CV↗
arXiv2026

Evolving Agent Interfaces

Computer-use agents are limited by the interfaces humans give them. An agent may type a spreadsheet cell by cell when a single line of code would do, or print all two thousand lines of a file through a shell to find one function. Both the interfaces and the knowledge of when to use each are built by hand, which cannot keep pace with new applications, workflows, and models. We argue that interface design should itself be an agent capability. We introduce Agent S4, an autonomous computer-use scientist that builds Evolving Agent Interfaces from task experience. A meta-agent examines task agents' trajectories on synthetic training tasks, without success labels, and constructs and revises interfaces and usage knowledge in a repository. Population-based evolution then selects which repositories to develop further by their performance on validation tasks the meta-agent never sees, favoring changes that generalize beyond the seen tasks. On OSWorld V2, Agent S4, with Qwen3.8-27B and the evolved repository of interfaces, sets a new state of the art in open-source computer-use agents, reaching 21.3% binary score, 3.8 times the same model with fixed interfaces. This is above both GPT-5.5 (13.0%) and Claude Opus 4.8 (20.6%), at about 4.3% and 1.5% of their respective costs. On ScienceBoard, the evolved interfaces improve the success rate from 41.4% to 58.6% with the same model. These results suggest that useful interfaces can be evolved from experience, making autonomous interface design a source of improvement complementary to advances in models and harnesses. We open-source our code at https://github.com/simular-ai/agent-s.

cs.AI↗
arXiv2026

Auditor-Assisted Summary-Channel Verification for Hosted LLM Identity Substitution

Hosted LLM services increasingly expose model identity as a product claim, but external users and auditors often cannot inspect which model configuration served a request. We study an auditor-assisted regime in which a trusted auditor enrolls claimed identities and prepares verification material before deployment, while audit-time verification uses only the claimed identity and released output, without runtime access to model internals, inference logs, routing decisions, or provider metadata. We propose SUMMARK, a summary-channel watermarking framework that uses user-visible reasoning summaries as a keyed audit surface. During setup, SUMMARK constructs identity-specific vocabulary carriers, trains lightweight summary adapters, and calibrates per-identity thresholds. At deployment time, a keyed statistical test assesses whether the released summary is consistent with the claimed identity and, when needed, attributes the most likely enrolled identity. Across model families, scales, and task domains, SUMMARK reliably detects cross-family and cross-scale substitutions at low false-positive rates while preserving task-facing answer quality. We further study FP16->INT8/INT4 precision-consistency diagnostics and robustness under bounded post-processing, rewriting stress tests, and text-only spoofing, revealing both strong performance and clear failure boundaries. Our code is publicly available at https://github.com/trinitr0toluene/Summark.

cs.CR↗
arXiv2026

Unified Sharp Bounds on Probabilities of Causation

Probabilities of causation (PoCs) have developed steadily, from the binary measures first introduced by Pearl to the non-binary setting recently studied by Shu et al. As the theory has matured, PoCs have found applications in a growing range of areas, including personalized medicine, marketing and legal attribution. Shu et al. proposed closed-form bounds for multi-valued treatments and outcomes and proved them valid, but left their sharpness in high dimensions open. We close this gap with a mathematical proof. We further show that $\PNS$, $\PSub$, $\PRep$ and $\PN$ are all special cases of a single, more general formula. Combined with their equivalence classes, this settles the question of sharp bounds for discrete PoCs from experimental and observational data: in every dimension, they are given in closed form by a single theorem.

cs.AI↗
arXiv2026

End-to-End Context Compression at Scale

Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length. Recent techniques to compress the KV cache fall short: they either degrade model quality substantially or require considerable time and compute to compress a single long prompt. Furthermore, many methods require the input to fit within the target model's context window, and are generally incompatible with modern production inference engines. Encoder-decoder compressors, which map a long token sequence to a shorter sequence of latent embeddings consumed by a decoder, are an appealing alternative in principle. However, existing approaches are not competitive with KV cache compression on the accuracy-efficiency frontier. In this work, we revisit encoder-decoder compression and close this gap. We first perform an architecture search, pre-training many variants from scratch to determine how best to design and train encoder-decoder compressors. Guided by our findings, we continually pre-train a family of 0.6B-encoder, 4B-decoder models on nearly 400B tokens each, at compression ratios of 1:4, 1:8, and 1:16. We introduce Latent Context Language Models (LCLMs), a family of compressors that improve the Pareto frontier across general-task performance, compression speed, and peak memory usage. LCLMs also unlock new efficient downstream applications: they serve as backbones for long-horizon agents that skim a compressed long context and adaptively expand relevant segments on demand, and they can compact the model's own generated reasoning into latent memory during inference, which is nearly lossless on the full MATH test set while being highly efficient, reducing peak KV usage by up to $7.9\times$.

cs.CL↗
arXiv2026

Frozen Models, Evolving Expertise: Model-Agnostic Learning from Deployment Experience for Multimodal Medical AI

Large language models (LLMs) and vision-language models (VLMs) are usually frozen after deployment, so they do not learn from the cases they solve. This is especially concerning in medicine, where new clinical evidence, updated guidelines, and new therapies can change established practice. Fine-tuning can update the model, but it requires access to model weights and additional training. Parameter-free methods avoid training, but they may overfit a fixed validation set, lack reliable domain knowledge, or lose visual details by saving experience only as text. To address these limitations, we present a model-agnostic framework that allows frozen LLMs and VLMs to learn from deployment experience through three forms of external expertise: a Skill that guides reasoning and tool use, a Knowledge Memory that stores reliable facts supported by earlier cases or trusted external evidence, and a Multimodal Knowledge Base that keeps visual examples and guides the model to relate each retrieved case to the current image. Instead of relying on a fixed validation set, a validation strategy keeps an update only if it helps on new cases without degrading performance on earlier ones. Across six benchmarks covering clinical diagnosis, clinical workflows, medical reasoning, and medical and non-medical visual reasoning, and with four open-weight and closed-source base models, our framework improves performance during online deployment by up to 34.2% over the base model on medical tasks, generalizes to unseen cases, transfers to other models without further optimization, and works in non-medical domains.

cs.AI↗
arXiv2026

LEASH: Uncertainty-Aware Weak-to-Strong Escalation and Oversight for Computer-Use Agents

Computer-use agents increasingly tackle long-horizon tasks, where running a frontier model at every step is expensive and a small open-weight model alone often fails. Weak-to-strong escalation offers a middle ground where a weak agent acts by default and hands control to a strong one only when needed. The core challenge is deciding when to escalate, which requires an uncertainty signal cheap enough to check at every step. However, even after handoff, the strong agent can still get lost, but a black-box API that the strong agent is usually served exposes no internals to monitor. To this end, we propose LEASH, a training-free framework that unleashes the strong agent only when the weak agent's latent uncertainty signals it is needed, and keeps it on a leash afterward. LEASH has three layers: (1) initial task uncertainty for early routing, (2) step-level uncertainty for runtime escalation, and, after handoff, (3) a shadow observer for continued oversight of the strong black-box agent. Here, we use latent uncertainty as a fuel for all three layers. Specifically, we compare the weak model's hidden states with those from its past successful and failed trajectories, and a state that looks more like the failures signals higher uncertainty. During weak-agent execution, our uncertainty estimator reuses activations already produced during action generation, so it needs no extra forward pass or dedicated monitor training. Experiments on OSWorld across four contemporary model backbones show that LEASH consistently improves over standalone weak agents and outperforms random escalation by 6.55--11.94 points at matched escalation rates. With Qwen3.5-35B-A3B, LEASH achieves 73.66% task success at $0.4117 per task, matching always-on Claude Opus 5 (72.90%) while reducing inference cost by 19.2%.

cs.AI↗
arXiv2026

MFSA: A Multi-Format Systolic Array with Native Micro Scaling Support

AI workloads increasingly employ low-precision formats such as BF16 / FP16, FP8 variants, FP4, Micro Scaling (MX), and INT modes across models, layers, and execution stages to balance accuracy, memory footprint, and throughput, which in turn requires hardware to support these formats in a flexible and efficient way. We present MFSA, a weight-stationary multi-format systolic array that achieves area efficiency through two architectural choices. At the PE level, MFSA uses dot-product-accumulator-based processing elements extended from an existing open-source trans-precision FPU to support reduced-precision formats (FP32/FP16/BF16/CFP8/FP4/INT16/INT8/INT4) with a shared datapath and proportional throughput scaling. At the array level, MFSA employs Online Block-Scale Rebasing (OBSR) for block-scaled formats, which folds each block scale into the running FP32 partial sum at block boundaries without adding a second cross-block accumulator, thereby reducing the additional overhead of block-scaling support. In TSMC 28 nm at 1 GHz, matched-throughput BF16/FP8/FP4 cascaded FMA-PE baselines are 1.5x/2.5x/4.5x larger than the single-format DPA-PE, the gap growing with subword parallelism. A routed 16x16 MFSA reaches a peak 0.5/1/2/4 TOPS of FP32/BF16/FP8/FP4, and OBSR cuts the incremental MX overhead by 64.2% in area and 67.5% in energy versus a naive MX implementation baseline, without hurting the numerical stability on real LLM traces. Adding support for all other formats based on a BF16+CFP8 systolic array costs 25.7% additional area. OBSR also extends to non-power-of-two block scales: folding the fractional part of an E4M3 scale into the MFPE dot product supports NVFP4 alongside OCP MX for 11.8% area and 7.3% power over a CFP8+FP4 base, less than half the cost of a naive implementation.

cs.AR↗
arXiv2026

Proton Structure from Neural Simulation-Based Inference at the LHC

The precise determination of the parton distribution functions (PDFs) of the proton is an essential ingredient for LHC analyses, including for those at the upcoming High-Luminosity LHC. So far, PDFs are determined from global fits to binned low-dimensional data obtained from unfolded hard-scattering cross section measurements. In this work we demonstrate for the first time the feasibility of neural simulation-based inference (NSBI) for constraining the proton PDFs using a high-dimensional unbinned data set. Exploiting the full statistical power of unbinned data removes the loss of information inherited by the binning procedure. As a proof-of-concept, we determine the gluon PDF from simulated data of top quark pair production at the LHC with $\sqrt{s}=13$ TeV. Taking into account both experimental and theoretical systematic uncertainties in the detector-level features, we demonstrate how the NSBI pipeline achieves significant improvements in precision compared to low-dimensional binned analyses. Our results illustrate the potential of unbinned inference to reduce the reliance on coarse approximations of uncertainties and their correlations entering PDF determinations, hence contributing to a new paradigm of unbinned detector-level ML-assisted measurements at the LHC.

hep-ph↗
arXiv2026

Finite-Temperature Equations of State for Neutron-Star Mergers within the Quark Mean-Field Model

We extend the quark mean-field (QMF) model to finite temperature and systematically investigate the thermodynamic properties of uniform nuclear matter, using the relativistic mean-field (RMF) model as a reference. For each model, we consider two parameter sets calibrated to the same selected nuclear-matter saturation properties but with different nucleon Dirac effective-mass ratios at saturation. We examine the effective nucleon mass and the composition of $β$-equilibrium matter, together with the free energy, energy density, pressure, entropy, and heat capacities of symmetric nuclear matter, pure neutron matter, and $β$-equilibrium matter. The QMF and RMF models exhibit similar qualitative trends, but their predictions increasingly differ at high density, where the effective masses have distinct density dependences. These differences are particularly evident in the thermal pressure and thermal index. For the two parameter sets considered in this work, we further provide compact analytical representations of the thermal index, $Γ_{\rm th}(ρ_{\rm b},T,β)$ over baryon densities of 0.08-1.0 $\rm fm^{-3}$, temperatures of 5-100 MeV, and isospin asymmetries of 0-1. The fits reproduce the numerical results with an accuracy better than $2\%$ and can be directly implemented in binary neutron-star merger simulations. This study establishes a finite-temperature extension of the QMF framework and provides a basis for future assessments of how a quark-structure-based description of nuclear matter affects thermal support and post-merger dynamics in binary neutron-star mergers.

nucl-th↗