Searcharxiv⌕ Search

arXiv subjects

Ang Li

Publications and source records attributed to Ang Li.

At least 73 records · Page 4Linked to original sources
arXiv2026

VOID: Defeating Unauthorized Mimicry in Latent Diffusion Models

While Latent Diffusion Models (LDMs) have revolutionized visual synthesis, they are increasingly exploited for unauthorized mimicry of individuals. Existing defenses inject deceptive perturbations to steer the generated images toward irrelevant targets. However, this approach hinges on an ungrounded assumption: subtle perturbations can maintain their deceptive efficacy throughout an LDM's extensive generation process. In reality, the model's innate restoration mechanism will remove such perturbations and cause individual identities to re-emerge in the images generated. We propose VOID, a defense framework that overcomes this conundrum by manipulating an LDM's intrinsic stochasticity. VOID perturbs the diffusion pipeline in two novel ways: 1) amplifying the latent encoding errors to shatter an image's semantic structure, and 2) counteracting the target guidance signals to suppress the model's restoration capabilities. This results in a semantic corruption that thwarts any unauthorized mimicry. Notably, the security gain does not come at the price of visual utility, as VOID simultaneously manages to confine perturbations to human-imperceptible regions of protected images. Our comprehensive evaluation of 24 state-of-the-art defenses against 10 mimicry attacks on 5 datasets demonstrates VOID's unprecedented protection power: it increases the average Frechet Inception Distance (FID) from 113 to 365, a 223% improvement over the strongest defense to date.

cs.CV↗
arXiv2026

Federated Large Language Models: Current Progress and Future Directions

Large Language Models have achieved impressive performance across diverse applications, yet their training typically depends on centralized data collection, raising serious privacy and governance concerns. Federated Learning offers a decentralized alternative by enabling multiple clients to collaboratively train shared models without exposing raw local data. However, integrating FL with LLMs introduces new challenges, including data heterogeneity, convergence instability, communication overhead, and computational constraints. This survey provides a comprehensive and up-to-date overview of Federated Learning for Large Language Models (FedLLM). We systematically review recent advances, with particular emphasis on federated fine-tuning and federated prompt learning, and analyze how existing methods address efficiency, personalization, and security challenges. We further summarize emerging directions such as federated pre-training and federated agents. Our goal is to offer a structured perspective on this rapidly evolving field and to highlight promising avenues for future research.

cs.LG↗
arXiv2026

MaterialClusterGS: Palette-Based Material Decomposition and Physically-Based Relighting with 2D Gaussian Splatting

We present MaterialClusterGS, a palette-based material decomposition framework for 2D Gaussian Splatting that enables physically based relighting and material editing. Existing Gaussian inverse rendering methods typically assign independent BRDF parameters to individual primitives. While flexible, this local fitting strategy makes material recovery highly under-constrained: shadows, indirect illumination, geometric errors, and visibility residuals can be absorbed into thousands of slightly different local material estimates. Meanwhile, recent palette-based appearance methods operate solely in RGB space without modeling physical materials or illumination. To bridge this gap, we represent scene materials using a compact global palette of shared BRDF prototypes assigned via a continuous spatial material field. Without shared material structure, editing one region does not propagate consistently to others of the same material, making per-primitive decompositions impractical for editing. We jointly optimize the material field, palette prototypes, and environment lighting under a physically based rendering objective. The resulting framework recovers compact, spatially coherent attributes directly usable for material editing, relighting, and transfer.

cs.GR↗
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 over 350B 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. We demonstrate that LCLMs serve as efficient backbones for long-horizon agents, letting the agent skim through a compressed long context and adaptively expand relevant segments on demand.

cs.CL↗
arXiv2026

Convolution-type Identity for Characteristic Polynomials of Geometric Semilattices

We establish a convolution formula for the characteristic polynomial of a finite geometric semilattice $M$: \[ χ(M,st)=\sum_{X\in \underline{M}} s^{r-{\rm rk}_{\underline{M}}(X)}χ(\underline{M}^X,t)\,χ(M_{(X)},s), \] where $\underline{M}$ denotes the centralization of $M$, and $M_{(X)}$ denotes the localization at $X$. This generalizes a nice formula of Southerland, Southern, and Zhou, which is recovered at $s=1$. When specialized to hyperplane arrangements, the identity yields a new expansion closely related to Wang's convolution formula. We further provide a combinatorial interpretation of the convolution formula using the finite field method over $\mathbb{F}_{p^2}$ and $\mathbb{F}_p$.

math.CO↗
arXiv2026

In-medium effects of nucleon-nucleon cross sections in heavy-ion collisions

Based on the isospin-dependent Boltzmann-Uehling-Uhlenbeck transport model, we systematically investigate the in-medium effects of nucleon-nucleon ($NN$) cross sections on nucleonic and pionic observables in heavy-ion collisions, employing microscopic cross sections derived from the Brueckner-Hartree-Fock approach. Key observables include nuclear stopping, the neutron-to-proton ($n/p$) ratio, neutron-proton transverse flow differences, differential collective flow, pion multiplicities, and the resulting $(π^-/π^+)_{\rm like}$ ratio. The analysis disentangles the respective contributions from the scattering amplitude, the density of states, and the total momentum ($K$) of the colliding pairs. We find that larger in-medium $NN$ cross sections generally enhance free nucleon emission and nuclear stopping, with the nucleon effective mass playing a dominant suppressive role. However, it is insufficient to account only for the medium corrections from effective mass: both the medium effect from the scattering amplitude and the $K$-dependence exert noticeable influences on the observables. In particular, nuclear stopping is found to be highly sensitive to these in-medium modifications of cross sections. While the $n/p$ ratio and transverse flow difference remain largely insensitive, the differential collective flow and pion yields are strongly affected. These results indicate that the interplay between scattering amplitude, density-of-states and $K$-dependence is essential to accurately describe medium effects in heavy-ion collisions.

nucl-th↗
arXiv2026

HiMed: Incentivizing Hindi Reasoning in Medical LLMs

Medical large language models hold promise for reducing healthcare disparities, yet Hindi remains severely underrepresented. While medical LLMs excel in high-resource languages, their performance degrades sharply in Hindi, particularly on Indian systems of medicine. We argue that robust cross-lingual medical transfer requires Hindi reasoning. To this end, we introduce HiMed, a Hindi reasoning medical corpus and benchmark suite covering both Western and Indian medicine. We further propose HiMed-8B, a Hindi-form medical reasoning LLM, through the design of decaying scaffolding reward. Extensive experiments demonstrate improvement in Hindi medical reasoning performance and reduction in the English--Hindi accuracy gap. Ablation studies validate the contribution of each training stage and reward component. All data and code are available on GitHub: https://github.com/FreedomIntelligence/HiMed.

cs.CL↗
arXiv2026

Transpiler-Architecture Co-Design to Curb Clifford Costs in Fault-Tolerant Quantum Computing

Quantum Error Correction (QEC) codes form the foundation of Fault-Tolerant Quantum Computing (FTQC) and predominantly use the Clifford+T gate set. Recently, Clifford operations have become the key performance bottleneck in implementing QEC. While state-of-the-art approaches like Pauli-Based Compilation (PBC) reduce Clifford overhead by transforming Clifford gates into Pauli measurements, they do so at the cost of gate-level parallelism, inflating circuit depth and execution times. To overcome these limitations, we introduce TACO, a Transpiler-Architecture Co-design framework that tackles the Clifford bottleneck through circuit and architectural optimization. TACO uses FTQC insights to guide hardware-aware Clifford gate elimination and circuit restructuring, and leverages the resulting optimized circuits to refine architectural design. TACO applies FTQC-specific transformations to aggressively reduce Clifford overhead from rotation synthesis and Toffoli decompositions, while preserving gate-level parallelism. The resulting architecture is optimized for the locality and data-movement patterns of these circuits, enabling high-throughput, resource-efficient execution. Our evaluation across diverse benchmarks shows that TACO achieves up to 21.9x (mean 4.4x) reduction in execution time compared to the state-of-the-art baseline.

quant-ph↗
arXiv2026

Multiwavelength Analysis of PSR J0437-4715 with Pulse Profile Modeling

We present a multi-wavelength analysis of the nearby millisecond pulsar PSR J0437--4715, combining Hubble Space Telescope (HST) far-ultraviolet, ROSAT soft X-ray, and XMM-Newton X-ray data, to model its broadband emission and energy-resolved pulse profiles, and infer key stellar parameters via Bayesian inference. The broadband emission includes cold thermal, hot thermal, and non-thermal components: cold bulk surface emission is modeled with a non-magnetized partially-ionized hydrogen atmosphere; hot-spot emission adopts the pulse profile modeling technique with a non-magnetized fully-ionized hydrogen atmosphere model; and non-thermal emission is included as a phase-invariant power-law component. By adopting an informative prior on the hot-spot geometry informed by radio polarization position angle measurements, the joint multi-instrument analysis yields a statistically viable and radio-consistent solution with a gravitational mass of 1.38$\pm$0.03~M$_\odot$ and an equatorial circumferential radius of 13.25$_{-0.35}^{+0.34}$~km (68\% confidence intervals). The hot-spot geometry consists of two spherical caps with uniform temperature distributions: the primary hot spot is situated at a colatitude of $\approx$130$^{\circ}$, and the secondary hot spot lies at a colatitude of $\approx$9$^{\circ}$, close to the north pole. It yields tighter radius constraints than HST+ROSAT fits and shifts the radius posterior distribution to larger values relative to NICER-only fits. This work demonstrates the importance of multi-wavelength data in refining neutron star mass-radius measurements and resolving geometric degeneracies.

astro-ph.HE↗
arXiv2026

Stateful Reasoning via Insight Replay

Chain-of-Thought (CoT) reasoning has become a foundation for eliciting multi-step reasoning in large language models, but recent studies show that its benefits do not scale monotonically with chain length: while longer CoT generally enables a model to tackle harder problems, on a given problem, accuracy typically increases with CoT length up to a point, after which it declines. We identify a major cause of this phenomenon: as the CoT grows, the model's attention to critical insights produced earlier in the trace gradually weakens, making those insights progressively less accessible when they are most needed. Therefore, we propose \textbf{InsightReplay}, a stateful reasoning approach in which the model periodically extracts critical insights from its reasoning trace and replays them near the active generation frontier, keeping them accessible as the reasoning scales. Extensive experiments on a $\mathbf{2}\!\times\!\mathbf{3}\!\times\!\mathbf{4}$ benchmark grid, covering model scales $\{\text{8B}, \text{30B}\}$, model families $\{\text{Qwen3.5}, \text{DeepSeek-R1-Distill-Qwen}, \text{Gemma-4}\}$, and reasoning benchmarks $\{\text{AIME}, \text{HMMT}, \text{GPQA Diamond}, \text{LiveCodeBench v5}\}$, show that 3-round InsightReplay yields accuracy gains across \textbf{all 24 settings}, with an averaged improvement of $\mathbf{+1.65}$ points over standard CoT, and a largest single-setting gain of $\mathbf{+9.2}$ points on R1-Distill-32B's LiveCodeBench v5 subset. Our results suggest that the effectiveness of test-time scaling depends not only on how much a model reasons, but also on whether critical intermediate insights remain accessible throughout long reasoning trajectories.

cs.AI↗
arXiv2026

Benchmarking and Resource Analysis for Augmented-Lagrangian Quantum Hamiltonian Descent

Quantum Hamiltonian Descent (QHD) is a continuous optimization algorithm based on simulating a time-dependent quantum Hamiltonian whose potential energy encodes the objective function and whose kinetic energy promotes exploration through quantum interference and tunneling. While QHD is formulated for unconstrained optimization, many real-world optimization problems are constrained and highly nonconvex. In this paper, we benchmark AL-QHD, a hybrid framework that embeds QHD within the Augmented Lagrangian Method (ALM), thereby solving a sequence of unconstrained subproblems while using ALM to enforce constraints. We evaluate AL-QHD on standard nonconvex test functions and use iterative refinement to improve solution accuracy at fixed per-run qubit cost. We also perform a gate-based resource analysis on ACOPF-derived power system subproblems constructed from power-network data to estimate the quantum-computer scale required for practical applications. Resource estimates on Texas7k-derived ACOPF instances show steep hard-gate scaling, reaching $\sim 4.46 \times 10^7$ entangling gates in a NISQ-oriented model and $\sim 9.42 \times 10^8$ T gates in a fault-tolerant model at $\sim 5.3 \times 10^2$ active variables. These results suggest that AL-QHD is a useful framework for studying constrained nonconvex optimization with QHD, but that practical ACOPF-scale applications would likely require large-scale fault-tolerant quantum hardware.

quant-ph↗
arXiv2026

Capacity-Aware Inference: Mitigating the Straggler Effect in Mixture of Experts

The Mixture of Experts (MoE) is an effective architecture for scaling large language models by leveraging sparse expert activation to balance performance and efficiency. However, under expert parallelism, MoE suffers from inference inefficiencies due to imbalanced token-to-expert assignment, where underloaded experts complete computations early but must wait for overloaded experts, leading to global delays. We define this phenomenon as the \textbf{\textit{Straggler Effect}}, as the most burdened experts dictate the overall inference latency. To address this, we first propose \textit{\textbf{Capacity-Aware Token Drop}}, which enforces expert capacity limits by discarding excess tokens from overloaded experts, effectively reducing load imbalance with minimal performance impact (e.g., $30\%$ speedup with only $0.9\%$ degradation on OLMoE). Next, given the presence of low-load experts remaining well below the capacity threshold, we introduce \textit{\textbf{Capacity-Aware Expanded Drop}}, which allows tokens to include additional local experts in their candidate set before enforcing strict local capacity constraints, thereby improving load balance and enhancing the utilization of underused experts. Extensive experiments on both language and multimodal MoE models demonstrate the effectiveness of our approach, yielding substantial gains in expert utilization, model performance, and inference efficiency, e.g., applying Expanded Drop to Mixtral-8$\times$7B-Instruct yields a {0.2\%} average performance improvement and a {1.85$\times$} inference speedup. The code is released at: https://github.com/CASE-Lab-UMD/Capacity-Aware-MoE.

cs.LG↗
arXiv2026

Demystifying When Pruning Works via Representation Hierarchies

Network pruning, which removes less important parameters or architectures, is often expected to improve efficiency while preserving performance. However, this expectation does not consistently hold across language tasks: pruned models can perform well on non-generative tasks but frequently fail in generative settings. To understand this discrepancy, we analyze network pruning from a representation-hierarchy perspective, decomposing the internal computation of language models into three sequential spaces: embedding (hidden representations), logit (pre-softmax outputs), and probability (post-softmax distributions). We find that representations in the embedding and logit spaces are largely robust to pruning-induced perturbations. However, the nonlinear transformation from logits to probabilities amplifies these deviations, which accumulate across time steps and lead to substantial degradation during generation. In contrast, the stability of the categorical-token probability subspace, together with the robustness of the embedding space, supports the effectiveness of pruning for non-generative tasks such as retrieval and multiple-choice selection. Our analysis disentangles the effects of pruning across tasks and provides practical guidance for its application. Code is available at https://github.com/CASE-Lab-UMD/Pruning-on-Representations

cs.CL↗
arXiv2026

DisagMoE: Computation-Communication overlapped MoE Training via Disaggregated AF-Pipe Parallelism

Mixture-of-experts (MoE) architectures enable trillion-parameter LLMs with sparsely activated experts. Expert parallelism (EP) is a widely adopted MoE training strategy, but it suffers from severe all-to-all communication bottlenecks, which is exaggerated by the limited inter-node network bandwidth as the growing model size requires distributing experts across GPU nodes. Prior work focused on overlapping these all-to-all communications with feed-forward network (FFN) and self-attention computations, which often leaves residual network-bound stalls due to inherent imbalance in attention and FFN layers' computation-communication ratios. We present DisagMoE, a disaggregated MoE training system that jointly optimizes model placement and scheduling for maximal efficiency. DisagMoE separates attention and FFN layers into disjoint GPU groups, introduces a multi-stage pipeline with uni-directional, many-to-many communications, and employs a computation-communication roofline model to balance GPU and network bandwidth allocation among the attention and FFN groups. DisagMoE is implemented on Megatron-LM, and evaluation shows that DisagMoE improves training efficiency across multiple MoE models with up to 1.8x speedup on 16-node 8xH800 clusters.

cs.LG↗
arXiv2026

TransDot: An Area-efficient Reconfigurable Floating-Point Unit for Trans-Precision Dot-Product Accumulation for FPGA AI Engines

Commercial FPGAs, such as AMD Versal devices, increasingly incorporate AI engines that exploit low-precision packed-SIMD fused multiply-accumulate (FMA) to achieve proportional throughput gains. However, trans-precision FMA (e.g., multiplying two FP16 numbers and adding their result to an FP32 accumulator), which preserves numerical stability by accumulating in higher precision, remains bottlenecked by the highest-precision, lowest-throughput operation. Dot-product accumulation (DPA) (e.g., performing a dot-product on two 4-element FP8 vectors and adding its result to an FP32 accumulator) can fully utilize the input/output bandwidth and computational resources. Existing flexible open-source FPUs, such as FPnew, do not support DPA and implement SIMD FMA on low-precision formats by replicating independent FMA lanes, which increases area, underutilizes shared arithmetic resources, and complicates the integration of DPA operations. This paper presents TransDot, a reconfigurable FPU that unifies multi-precision SIMD FMA and trans-precision DPA within a shared, reconfigurable datapath. TransDot extends the baseline design with 2-term FP16, 4-term FP8, and 8-term FP4 dot-product accumulation into FP32 using reconfigurable subcomponents. Evaluation shows that TransDot delivers 2$\times$ FP16, 4$\times$ FP8, and 8$\times$ FP4 throughput via DPA with FP32 accumulation, and 1.46$\times$ area efficiency in FP16 DPA and 2.92$\times$ area efficiency in FP8 DPA, at the cost of 37.3% larger area on average and an additional pipeline stage in dot-product mode compared to the FPnew baseline. These results demonstrate that TransDot's area-efficient design enables scalable deployment in next-generation AMD Versal AI engines.

cs.AR↗
arXiv2026

Beyond Autoregressive RTG: Conditioning via Injection Outside Sequential Modeling in Decision Transformer

Decision Transformer (DT) formulates offline reinforcement learning as autoregressive sequence modeling, achieving promising results by predicting actions from a sequence of Return-to-Go (RTG), state, and action tokens. However, RTG is a scalar that summarizes future rewards, containing far less information than typical state or action vectors, yet it consumes the same computational budget per token. Worse, the self-attention cost of Transformers grows quadratically with sequence length, so including RTG as a separate token adds unnecessary overhead. We propose SlimDT, which removes RTG from the autoregressive sequence. Instead, we inject RTG information into the state representations before the sequential modeling step, allowing the Transformer to process only a compact (state, action) sequence. This reduces the sequence length by one-third, directly improving inference efficiency. On the D4RL benchmark, SlimDT surpasses standard DT across various tasks and achieves performance comparable to existing state-of-the-art methods. Decoupling a sparse conditioning signal from an information-rich sequence thus yields both computational gains and higher task performance.

cs.LG↗
arXiv2026

DICE: Enabling Efficient General-Purpose SIMT Execution with Statically Scheduled Coarse-Grained Reconfigurable Arrays

While GPUs dominate massively parallel computing through the single-instruction, multiple-thread (SIMT) programming model, their underlying single-instruction, multiple-data (SIMD) execution incurs substantial energy overhead from frequent register file (RF) accesses and complex control logic. We present DICE, a novel architecture that addresses these inefficiencies by replacing the SIMD backend with minimal-overhead, statically scheduled coarse-grained reconfigurable arrays (CGRAs). Unlike SIMD units that execute warps of threads in lockstep, DICE dispatches active threads in a pipelined manner onto the CGRA fabric, where data flow directly between processing elements (PEs), reducing RF accesses for intermediate values. To handle operations with runtime dynamism, such as variable-latency memory loads and data-dependent control flow, while preserving static scheduling, DICE compiles programs into "p-graphs" by partitioning dynamic dependence edges across separate CGRA configurations. DICE further introduces several key optimizations: double-buffered configuration memory to hide reconfiguration latency, compile-time p-graph unrolling to enhance resource utilization, and a temporal memory coalescing unit (TMCU) to merge memory requests from consecutive, pipelined threads. Evaluations on Rodinia benchmarks in Accel-sim demonstrate that DICE reduces register file accesses by 68% on average. With equivalent computation and memory resources, DICE's CGRA Processors (CPs) achieve a geometric mean of 1.77-1.90x dynamic energy efficiency and 42.0%-45.9% average power reduction compared to the modeled NVIDIA Turing Streaming Multiprocessors (SMs), while the full DICE system achieves performance comparable to the modeled Turing GPU baselines. DICE demonstrates that spatial pipeline execution can deliver substantial energy savings without sacrificing performance.

cs.AR↗
arXiv2026

ISExplore:Informative Segment Selection for Efficient Personalized 3D Talking Face Generation

Talking Face Generation (TFG) methods based on Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS) have recently achieved impressive progress in personalized talking head synthesis. However, existing methods typically require several minutes of reference video for meticulous preprocessing and fitting, resulting in hours of preparation time and limiting their practical applicability. In this paper, we revisit a fundamental yet underexplored question: do high-quality personalized TFG models truly require minutes-long reference videos? Our exploratory study reveals that a carefully selected reference segment of only a few seconds can often achieve performance comparable to that of using the full reference video. This finding suggests that the informativeness of reference data is more critical than its duration. Motivated by this observation, we propose ISExplore (Informative Segment Explore), a simple yet effective segment selection strategy that automatically identifies the most informative short reference segment based on three key data quality dimensions: audio feature diversity, lip movement amplitude, and viewpoint diversity. Extensive experiments demonstrate that ISExplore reduces data processing and training time by over 5x for both NeRF- and 3DGS-based methods, while preserving high-fidelity generation quality. Our method provides a practical and efficient solution for personalized TFG and offers new insights into data efficiency in 3D talking face generation.

cs.CV↗