Searcharxiv⌕ Search

arXiv subjects

Jiajie Li

Publications and source records attributed to Jiajie Li.

At least 19 recordsLinked to original sources
arXiv2026

Programming AMD XDNA NPUs with Open-source Compiler Tools: A FlashAttention Case Study

Spatial NPUs such as AMD XDNA place compute tiles beside small local memories and leave data movement between them to software. Mapping a multi-stage workload onto such a device is largely a question of where the intermediate tensors live. We report what we learned making those choices for FlashAttention with the open-source IRON and MLIR-AIR flows. We compare four reference designs on XDNA 1 and XDNA 2: one runs each operator separately, two stream between operators on chip, and one fuses all three attention stages into a single kernel. The fused kernel holds the $\boldsymbol{QK}^{\mathsf T}$ scores in compute-tile local memory and reduces partial results over the cascade interconnect, so the scores never return to shared MemTile memory. On XDNA 2, it reaches 3.62 TFLOP/s over complete end-to-end execution, twice the IRON design, with 5.3 to 7.2 times the energy efficiency of the integrated GPU on the same chip at 2K tokens and above. It covers twelve LLM configurations, from BERT to DeepSeek, up to 128K tokens. Roofline analysis at each memory level explains this result and shows when to stop. XDNA 1 has lower ridge points, so streaming on chip already reaches the compute-bound regime: the same fusion that doubles throughput on XDNA 2 is nearly wasted on XDNA 1. Comparing a mapping's operational intensity against each level's ridge point predicts which case applies before writing any code. Fuse until the mapping clears that ridge point, then stop. We release the reference designs as maintained open source.

cs.AR↗
arXiv2026

Density Evolution of Soft-Decision Collapsed Projection-Aggregation Decoding for Reed-Muller Codes over the BIAWGN Channel

Reed-Muller (RM) codes have been shown to achieve capacity over a range of channels, and recently proposed projection-aggregation (PA) decoding has been experimentally shown to achieve near-maximum-likelihood decoding performance. These recent achievements motivate theoretical research on PA decoding. In this work, we analyze the density function of the soft output from collapsed projection-aggregation (CPA) decoding for RM codes over the binary-input additive white Gaussian noise (BIAWGN) channel. We prove that soft-decision CPA decoding returns an exact marginal probability and is symmetric. Based on the analysis, we build a density evolution model for CPA decoding. To simplify the density evolution, we approximate the projection and the fast Hadamard transform decoding using hard-decision decoding. Simulation results over the BIAWGN channel show that our proposed density evolution model captures the fast reduction in the mean and the variance of the soft information returned from the CPA decoding, which qualitatively explains the decoding mechanism and the fast convergence speed of the CPA decoding. We perform an asymptotic analysis based on the proposed density evolution, and we show that CPA decoding can achieve a vanishing error probability for RM codes with a vanishing code rate.

cs.IT↗
arXiv2026

LIME: Learning Intent-aware Camera Motion from Egocentric Video

Autonomous robots often need to move their camera before they can act: to inspect an object, reveal an occluded region, or obtain a view that responds to a user's intent. While vision-language navigation translates instructions to base motion and vision-language-action policies map instructions to manipulation actions, language-conditioned camera motion remains comparatively underexplored as a first-class action. We formulate language-conditioned camera motion generation: given a current RGB observation and a free-form natural-language intent, predict a relative target camera pose for the next observation. This task is inherently non-trivial: viewpoint changes are driven by latent perceptual intentions, and a valid motion may operate at different semantic granularity, from entering a room to looking around a corner, inspecting a visible object, or revealing an occluded detail. To model this structure, we mine multi-intention camera-motion supervision from egocentric video, pairing plausible intents and observation-gain descriptions with relative SE(3) target poses. We propose LIME, a vision-language camera-motion generator that combines an auto-regressive observation-gain output with a continuous flow-matching pose head. This design lets the model jointly predict what the next view should reveal while representing multi-hypothesis target views. Across experiments and downstream robotic tasks, we show that LIME can learn to actively choose camera poses from passive human video, turning ordinary egocentric recordings into supervision for intent-aware active perception.

cs.RO↗
arXiv2026

From Human Guidance to Autonomy: Agent Skill System for End-to-End LLM Deployment on Spatial NPUs

Spatial neural processing units (NPUs) provide an energy-efficient platform for edge LLM inference, but efficiently deploying an LLM end-to-end on such hardware remains labor-intensive. Although AI coding agents have begun to lower this cost, existing studies have largely focused on single-kernel optimization rather than end-to-end LLM deployment on resource-constrained spatial NPUs. We present a two-stage methodology, instantiated on the AMD XDNA 2 NPU, that progresses from human-guided development to agent autonomy. In the first stage, we develop a reference deployment of Llama-3.2-1B through human-guided agent assistance. The resulting implementation achieves a speedup of 2.2x on prefill and 4.0x on decode over the hand-optimized baseline, with the optimization trajectory and its lessons recorded as structured documentation throughout. In the second stage, we distill the documentation into an agent skill system consisting of eight phases, orchestrating the optimization and debugging skill sets, with numerical correctness strictly enforced at each phase. Using our agent skill system, we autonomously deploy eight additional decoder-only LLMs (Llama-3.2-3B, SmolLM2-1.7B, Qwen2.5-{0.5B, 1.5B, 3B}, Qwen3-{0.6B, 1.7B, 4B}) end-to-end on the AMD XDNA 2 NPU using the open-source compiler stack. To our knowledge, these models have not previously been deployed on AMD NPUs via any open-source software stack. Each deployment completes in 0.5-4 hours of agent wall time with almost no human guidance, and passes the numerical-correctness gates, demonstrating functional generalization to previously unencountered LLMs. Three of the eight match or exceed the sustained performance of our Llama-3.2-1B reference deployment, suggesting that the resulting implementations can be competitive without additional model-specific human engineering.

cs.LG↗
arXiv2026

FCUS-rPPG: A Fast-Converging Unsupervised Framework for Remote Photoplethysmography via Gradient Oscillation Suppression

Remote photoplethysmography (rPPG) enables non-contact extraction of blood volume pulse (BVP) signals using consumer-grade cameras. Recent unsupervised rPPG methods learn BVP representations without requiring ground-truth physiological annotations, yet their optimization is often hindered by noisy and unstable gradients, resulting in slow convergence and limited cross-domain generalization. In this paper, we propose FCUS-rPPG, a fast-converging unsupervised rPPG framework with strong generalization capability. Motivated by the observation that BVP representations exhibit both multi-spectral covariation and low-dimensional manifold structure, we design a spectrally shared backbone that facilitates BVP feature disentanglement while improving optimization efficiency. To jointly enhance convergence stability and generalization performance, we further develop a unified optimization framework operating at the gradient, loss-landscape, and feature-representation levels. Specifically, a post-verification masking mechanism filters out misleading gradients according to the weak-amplitude physiological prior of BVP signals; a perturbation-based loss landscape smoothing strategy steers optimization toward more generalizable flat minima; and a noise-aware null-space regularization constrains feature updates to the orthogonal complement of the noise subspace, thereby mitigating noise-induced representation drift. Extensive experiments on five datasets demonstrate that FCUS-rPPG requires only one training epoch, whereas existing methods typically require tens to hundreds of epochs. Notably, FCUS-rPPG consistently achieves state-of-the-art (SOTA) performance in cross-dataset evaluations. This study provides an efficient and robust solution to the real-world deployment of unsupervised rPPG. The source code will be publicly available at https://github.com/JiaJieLee/FCUS-rPPG.

cs.CV↗
arXiv2026

Rapid GPU-Based Pangenome Graph Layout

Computational Pangenomics is an emerging field that studies genetic variation using a graph structure encompassing multiple genomes. Visualizing pangenome graphs is vital for understanding genome diversity. Yet, handling large graphs can be challenging due to the high computational demands of the graph layout process. In this work, we conduct a thorough performance characterization of a state-of-the-art pangenome graph layout algorithm, revealing significant data-level parallelism, which makes GPUs a promising option for compute acceleration. However, irregular data access and the algorithm's memory-bound nature present significant hurdles. To overcome these challenges, we develop a solution implementing three key optimizations: a cache-friendly data layout, coalesced random states, and warp merging. Additionally, we propose a quantitative metric for scalable evaluation of pangenome layout quality. Evaluated on 24 human whole-chromosome pangenomes, our GPU-based solution achieves a 57.3x speedup over the state-of-the-art multithreaded CPU baseline without layout quality loss, reducing execution time from hours to minutes.

cs.DC↗
arXiv2026

Enabling Fast Polar SC Decoding with IR-HARQ

To extend the applications of polar codes within next-generation wireless communication systems, it is essential to incorporate support for Incremental Redundancy (IR) Hybrid Automatic Repeat Request (HARQ) schemes. For very high-throughput applications, Successive Cancellation (SC) decoding is particularly appealing for polar codes owing to its high area efficiency. In this paper, we propose modifications to SC decoders that employ special nodes to accelerate decoding. Our modifications enable the use of polar IR-HARQ with SC decoding for high throughput applications. Compared to the unmodified SC IR-HARQ scheme, our proposed approach allows us to achieve a 72% reduction in node traversals with a polar code of length 2048. Simulation results confirm that the proposed special node modifications do not cause any degradation in FER performance.

eess.SP↗
arXiv2026

Enhanced Successive Cancellation List Decoder for Long Polar Codes Targeting Air Interface

Polar codes are the first codes with a proven capacity-achieving capability, but their decoding faces several challenges, especially under long code lengths. In this paper, we target algorithmic improvements and analyses to enable the implementation of long polar codes (e.g., length 8K bits) by addressing key challenges in memory usage and computational complexity presented by successive cancellation list (SCL) polar decoding. Perturbation-enhanced (PE) SCL decoders with a list size of $L$ reach the decoding performance of the SCL decoder with a list size of $2L$. The proposed bias-enhanced (BE) SCL decoders, which simplify the PE SCL decoder based on insights gained by an ablation study, return similar decoding performance to PE SCL decoders. Also, proposed BE generalized partitioned SCL (GPSCL) decoders with a list size of $8$ have a $67\%$ reduction in the memory usage and similar decoding performance compared to SCL decoders with a list size of $16$, and it demonstrates that an accurate bias can be generated under a reduced number of codewords from the list and reduces the overhead from $\left(L-1\right)n$ XOR gates plus $n$ priority encoders to $n$ XOR gates, where $n$ is the code length. Furthermore, input-distribution-aware (IDA) decoding is applied to BE GPSCL decoders, which shows how an accurate bias is generated under a low-complexity decoder. Up to $5.4\times$ reduction in the computational complexity is achieved compared to SCL decoders with a list size of $16$, and negligible latency overhead is added to the decoding process. The degraded decoding performance is at most $0.05\text{ dB}$ compared to BE GPSCL decoders without IDA decoding. Lastly, we theoretically prove that the bias in the BE SCL decoder moves the received soft information toward valid polar codewords with a high likelihood, and explain the decoding performance gain.

cs.IT↗
arXiv2026

A Universal Systematic Method to Generate Error Patterns on Memoryless Channels

The high computational cost of approaching the performance of Maximum-likelihood (ML) decoding has limited its practical use for decades. Because the complexity grows exponentially with the message length, researchers have spent years developing algorithms like Ordered Statistics Decoding (OSD), Partial Ordered Statistics Decoding (POSD) and Guessing Random Additive Noise decoding (GRAND) which try to approach ML performance. OSD, POSD and GRAND work by trying to guess the error patterns affecting the received signals. However, there does not exist a systematic method to extend the error pattern guesses to novel channels. This work introduces a systematic method that uses the Probability Density Function (PDF) of a memoryless channel to generate a set of error patterns that can be applied on any future received signal on this channel. Simulation results show that our proposed method applied on GRAND, OSD and POSD generally matches or outperforms current pre-generated error patterns on additive white Gaussian noise (AWGN) channel, mixture of Gaussian distribution channels, Rayleigh fading channel with perfect knowledge of Channel State Information (CSI) and Rayleigh fading channel with no perfect knowledge of Channel State Information (NCSI).

eess.SP↗
arXiv2026

Chain-of-Adaptation: Surgical Vision-Language Adaptation with Reinforcement Learning

Conventional fine-tuning on domain-specific datasets can inadvertently alter a model's pretrained multimodal priors, leading to reduced generalization. To address this, we propose Chain-of-Adaptation (CoA), an adaptation framework designed to integrate domain knowledge while maintaining the model's inherent reasoning and perceptual capabilities. CoA introduces a structured reasoning format that enhances domain alignment without sacrificing general multimodal competence by reinforcement learning. Experiments on standard surgical benchmarks, under both in-distribution and out-of-distribution settings, demonstrate that CoA achieves higher accuracy, stronger generalization, and more stable behavior than supervised fine-tuning. Furthermore, ablation studies confirm that CoA effectively preserves the model's core visual-language abilities, providing a reliable pathway for domain specialization in VLMs.

cs.CV↗
arXiv2026

Selective Prior Synchronization via SYNC Loss

Prediction under uncertainty is a critical requirement for the deep neural network to succeed responsibly. This paper focuses on selective prediction, which allows DNNs to make informed decisions about when to predict or abstain based on the uncertainty level of their predictions. Current methods are either ad-hoc such as SelectiveNet, focusing on how to modify the network architecture or objective function, or post-hoc such as softmax response, achieving selective prediction through analyzing the model's probabilistic outputs. We observe that post-hoc methods implicitly generate uncertainty information, termed the selective prior, which has traditionally been used only during inference. We argue that the selective prior provided by the selection mechanism is equally vital during the training stage. Therefore, we propose the SYNC loss which introduces a novel integration of ad-hoc and post-hoc method. Specifically, our approach incorporates the softmax response into the training process of SelectiveNet, enhancing its selective prediction capabilities by examining the selective prior. Evaluated across various datasets, including CIFAR-100, ImageNet-100, and Stanford Cars, our method not only enhances the model's generalization capabilities but also surpasses previous works in selective prediction performance, and sets new benchmarks for state-of-the-art performance.

cs.CV↗
arXiv2026

An inertial minimal-deformation-rate framework for shape optimization

We propose a robust numerical framework for PDE-constrained shape optimization and Willmore-driven surface hole filling. To address two central challenges -- slow progress in flat energy landscapes, which can trigger premature stagnation at suboptimal configurations, and mesh deterioration during geometric evolution -- we couple a second-order inertial flow with a minimal-deformation-rate (MDR) mesh motion strategy. This coupling accelerates convergence while preserving mesh quality and thus avoids remeshing. To further enhance robustness for non-smooth or non-convex initial geometries, we incorporate surface-diffusion regularization within the Barrett-Garcke-N"urnberg (BGN) framework. Moreover, we extend the inertial MDR methodology to Willmore-type surface hole filling, enabling high-order smooth reconstructions even from incompatible initial data. Numerical experiments demonstrate markedly faster convergence to lower original objective values, together with consistently superior mesh preservation throughout the evolution.

math.NA↗
arXiv2026

HeuriGym: An Agentic Benchmark for LLM-Crafted Heuristics in Combinatorial Optimization

While Large Language Models (LLMs) have demonstrated significant advancements in reasoning and agent-based problem-solving, current evaluation methodologies fail to adequately assess their capabilities: existing benchmarks either rely on closed-ended questions prone to saturation and memorization, or subjective comparisons that lack consistency and rigor. In this work, we introduce HeuriGym, an agentic framework designed for evaluating heuristic algorithms generated by LLMs for combinatorial optimization problems, characterized by clearly defined objectives and expansive solution spaces. HeuriGym empowers LLMs to propose heuristics, receive evaluative feedback via code execution, and iteratively refine their solutions. We evaluate nine state-of-the-art models on nine problems across domains such as computer systems, logistics, and biology, exposing persistent limitations in tool use, planning, and adaptive reasoning. To quantify performance, we propose the Quality-Yield Index (QYI), a metric that captures both solution pass rate and quality. Even top models like GPT-o4-mini-high and Gemini-2.5-Pro attain QYI scores of only 0.6, well below the expert baseline of 1. Our open-source benchmark aims to guide the development of LLMs toward more effective and realistic problem-solving in scientific and engineering domains.

cs.LG↗
arXiv2025

Maximization of Supercapacitor Storage via Topology Optimization of Electrode Structures

As widely used electrochemical storage devices, supercapacitors deliver higher power density than batteries, but suffer from significantly lower energy density. In this work, we propose a topology optimization model for electrode structure to maximize energy storage in supercapacitors. The existence of minimizers to the resulting optimal control problem, which is constrained by a modified steady-state Poisson--Nernst--Planck system describing ionic electrodiffusion, has been theoretically established by using the direct method in the calculus of variation. Sensitivity analysis of the topology optimization model is performed to derive variational derivatives and corresponding adjoint equations. A gradient flow formulation discretized by a stabilized semi-implicit scheme is developed to solve the resulting topology optimization problem. Extensive numerical experiments present various porous electrode structures that own large area of electrode-electrolyte interface, demonstrating the effectiveness and robustness of the proposed topology optimization model and corresponding algorithm.

math.OC↗
arXiv2025

HF-VTON: High-Fidelity Virtual Try-On via Consistent Geometric and Semantic Alignment

Virtual try-on technology has become increasingly important in the fashion and retail industries, enabling the generation of high-fidelity garment images that adapt seamlessly to target human models. While existing methods have achieved notable progress, they still face significant challenges in maintaining consistency across different poses. Specifically, geometric distortions lead to a lack of spatial consistency, mismatches in garment structure and texture across poses result in semantic inconsistency, and the loss or distortion of fine-grained details diminishes visual fidelity. To address these challenges, we propose HF-VTON, a novel framework that ensures high-fidelity virtual try-on performance across diverse poses. HF-VTON consists of three key modules: (1) the Appearance-Preserving Warp Alignment Module (APWAM), which aligns garments to human poses, addressing geometric deformations and ensuring spatial consistency; (2) the Semantic Representation and Comprehension Module (SRCM), which captures fine-grained garment attributes and multi-pose data to enhance semantic representation, maintaining structural, textural, and pattern consistency; and (3) the Multimodal Prior-Guided Appearance Generation Module (MPAGM), which integrates multimodal features and prior knowledge from pre-trained models to optimize appearance generation, ensuring both semantic and geometric consistency. Additionally, to overcome data limitations in existing benchmarks, we introduce the SAMP-VTONS dataset, featuring multi-pose pairs and rich textual annotations for a more comprehensive evaluation. Experimental results demonstrate that HF-VTON outperforms state-of-the-art methods on both VITON-HD and SAMP-VTONS, excelling in visual fidelity, semantic consistency, and detail preservation.

cs.CV↗
arXiv2025

Topology optimization for microfluidic mixers by a phase field method

We investigate multi-physical topology optimization for microfluidic mixers employing the phase-field model. The optimization problem is formulated using a modified Ginzburg-Landau free energy functional. To eliminate fluid blockage in microfluidic mixers, we incorporate the coupled Navier-Stokes, convection-diffusion and Poisson-Boltzmann equations. An Allen-Cahn type gradient flow method is proposed based on sensitivity analysis. The algorithm is validated for its computational effectiveness through numerical simulations of benchmark problems in 2D and 3D.

math.OC↗
arXiv2025

Auto-Prompt Ensemble for LLM Judge

We present a novel framework that improves the reliability of LLM judges by selectively augmenting LLM with auxiliary evaluation dimensions. Existing LLM judges often miss crucial evaluation dimensions because they fail to recognize the implicit standards underlying human assessments. To address this challenge, we propose the Auto-Prompt Ensemble (APE), an adaptive framework that automatically learns evaluation dimensions from its failure cases. APE incorporates a confidence-based ensemble mechanism to decide when to adopt the judgments from additional evaluation dimensions through a novel confidence estimation approach called Collective Confidence. Extensive experiments demonstrate that APE improves the reliability of LLM Judge across diverse standard benchmarks. For instance, APE enhances GPT-4o agreement rate on Reward Bench from 87.2% to 90.5% in the zero-shot setting. Overall, APE provides a principled approach for LLM Judge to leverage test-time computation, and bridge the evaluation gap between human and LLM judges.

cs.AI↗
arXiv2025

Dato: A Task-Based Programming Model for Dataflow Accelerators

Recent deep learning workloads increasingly push computational demand beyond what current memory systems can sustain, with many kernels stalling on data movement rather than computation. While modern dataflow accelerators incorporate on-chip streaming to mitigate off-chip bandwidth limitations, existing programming models struggle to harness these capabilities effectively. Low-level interfaces provide fine-grained control but impose significant development overhead, whereas high-level tile-based languages abstract away communication details, restricting optimization and forcing compilers to reconstruct the intended dataflow. We present Dato, a Python-embedded, task-based programming model for dataflow accelerators that elevates data communication and sharding to first-class type constructs. Developers write programs as a graph of tasks connected via explicit stream types, with sharded inputs specified using layout types. These tasks are first mapped virtually onto the accelerator's spatial fabric, and the compiler then generates a physical mapping that respects hardware constraints. Experimental results on both AMD Ryzen AI NPU and Alveo FPGA devices demonstrate that Dato achieves high performance while significantly reducing the burden of writing optimized code. On the NPU, Dato attains up to 84% hardware utilization for GEMM and delivers a 2.81x speedup on attention kernels compared to a state-of-the-art commercial framework. On the FPGA, Dato surpasses leading frameworks in performance when generating custom systolic arrays, achieving 98% of the theoretical peak performance.

cs.PL↗