arXiv subjects

Wei Le

Publications and source records attributed to Wei Le.

At least 19 recordsLinked to original sources

arXiv2026

bioMoR: Biology-Guided Mixture-of-Recursions for Effective Genomic Learning

Transformer models for high-dimensional omics analysis process thousands of genes or pathways, although only a subset requires deep computation. Mixture-of-Recursions (MoR) improves efficiency through adaptive token-choice or expert-choice routing. We propose bioMoR, which, to the best of our knowledge, is the first framework to apply MoR to gene-level and pathway-level learning. Our contributions include identifying three locations for integrating structured biological knowledge within an MoR backbone: graph-based information sharing refines token embeddings, a structural bias guides self-attention toward biologically related tokens, and a graph-aware router uses neighborhood information to determine each token's recursion depth. These techniques are centered on our insight that additional knowledge of token interaction can effectively help models construct embeddings and select which tokens should be learned more deeply. Across eight benchmarks spanning diverse omics data types and evaluated under a unified five-fold cross-validation protocol, bioMoR improves average macro-F1 by 8.2 percentage points and balanced accuracy by 7.1 percentage points over the strongest biology-agnostic MoR baseline while using 75 percent fewer parameters and up to 58 percent fewer FLOPs than a non-recursive Transformer. The selected marker genes or pathways provide biological interpretability, while their token-specific recursion depths reveal how computation is allocated.

cs.AI↗

arXiv2026

BRITE: A Benchmark for Reliable and Interpretable T2V Evaluation on Implausible Scenarios

The rapid advancement of photorealistic Text-to-Video (T2V) generation brings in an urgent need for up-to-date evaluation methods. Existing benchmarks largely overlooked implausible scenarios and do not measure audio-visual alignment. We introduce BRITE, the first framework that unifies (1) implausible prompting, (2) fine-grained assessment of audio-visual consistency, and (3) QA-based interpretable evaluation into a comprehensive T2V benchmark. Unlike fully automated Multimodal LLM-based pipelines, which are prone to hallucination and prompt ambiguity, BRITE guarantees reliability through a rigorous human-in-the-loop protocol for benchmark creation. Evaluating five state-of-the-art models (Sora 2, Veo 3.1, Runway Gen4.5, Pixverse V5.5, and Qwen3Max), we reveal a critical performance gap: while models excel at static object composition, they exhibit significant degradation in object-action binding and audio-visual synchronization. Our framework offers the community a reliable, interpretable benchmark and evaluation framework that can detect and locate limitations in the next generation of T2V models, especially for off-manifold prompts

cs.MM↗

arXiv2026

Graph Transformer-Based Pathway Embedding for Cancer Prognosis

Accurate prediction of cancer progression remains a challenge due to the high heterogeneity of molecular omics data across patients. While biologically informed models have improved the interpretability of these predictions, a persistent limitation lies in how they encode individual genes to construct pathway representations. Existing hierarchical models typically derive gene features by directly mapping raw molecular inputs, whereas integration frameworks often rely on simple statistical aggregations of patient-level signals. These approaches often fail to explicitly learn a shared base representation for each gene, thereby limiting the expressiveness and biological accuracy of downstream pathway embeddings. To address this, we introduce PATH, a modulation-based, patient-conditioned gene embedding strategy. PATH represents a paradigm shift by starting from a shared base embedding for each gene, preserving a stable biological identity across the population, and then dynamically adapting it using patient-specific copy number variation (CNV) and mutation signals. This allows the model to capture subtle individual molecular variations while maintaining a consistent latent understanding of the gene itself. We integrate PATH into a graph transformer framework that models interactions among biologically connected pathways through pathway-guided attention. Across pancancer metastasis prediction, PATH achieves an F1 score of 0.8766, representing an 8.8 percent improvement over the current SOTA multi-omics benchmarks. Beyond superior predictive accuracy, our approach identifies biologically meaningful pathways and, crucially, reveals disease-state-specific pathway rewiring, offering new insights into the evolving pathway-pathway interactions that drive cancer progression.

cs.LG↗

arXiv2026

ConceptCoder: Improve Code Reasoning via Concept Learning

Large language models (LLMs) have shown promising results for software engineering applications, but still struggle with code reasoning tasks such as vulnerability detection (VD). We introduce ConceptCoder, a fine-tuning method that simulates human code inspection: models are trained to first recognize code concepts and then perform reasoning on top of these concepts. In prior work, concepts are extracted by multimodal models or LLMs to explain vision and natural language models. Our work is the first to formulate concepts for code. We define code concepts as human-understandable semantic properties of code and train models to learn such concepts. Our evaluation shows that this approach significantly improves VD accuracy, from 66.32 to 72.15 F1 on average over 9 open-source LLMs. ConceptCoder achieves the best VD performance compared to state-of-the-art (SOTA) baselines, including fine-tuned SOTA open-source LLMs and prompted proprietary models such as GPT-5.2 and Claude-Opus-4.5. Our approach also scales: concepts defined from four types of vulnerabilities benefit general vulnerability datasets with 134 CWEs. We further demonstrate that concept-based fine-tuning generalizes beyond VD and improves branch prediction. We release our code and datasets at https://figshare.com/s/1decab8232c653b44f71.

cs.SE↗

arXiv2025

CodeSense: a Real-World Benchmark and Dataset for Code Semantic Reasoning

Understanding and reasoning about code semantics is essential for enhancing code LLMs' abilities to solve real-world software engineering (SE) tasks. Although several code reasoning benchmarks exist, most rely on synthetic datasets or educational coding problems and focus on coarse-grained reasoning tasks such as input/output prediction, limiting their effectiveness in evaluating LLMs in practical SE contexts. To bridge this gap, we propose CodeSense, the first benchmark that makes available a spectrum of fine-grained code reasoning tasks concerned with the software engineering of real-world code. We collected Python, C and Java software projects from real-world repositories. We executed tests from these repositories, collected their execution traces, and constructed a ground truth dataset for fine-grained semantic reasoning tasks. We then performed comprehensive evaluations on state-of-the-art LLMs. Our results show a clear performance gap for the models to handle fine-grained reasoning tasks. Although prompting techniques such as chain-of-thought and in-context learning helped, the lack of code semantics in LLMs fundamentally limits models' capabilities of code reasoning. Besides dataset, benchmark and evaluation, our work produced an execution tracing framework and tool set that make it easy to collect ground truth for fine-grained SE reasoning tasks, offering a strong basis for future benchmark construction and model post training. Our code and data are located at https://codesense-bench.github.io/.

cs.SE↗

arXiv2025

Automatically Detecting Numerical Instability in Machine Learning Applications via Soft Assertions

Machine learning (ML) applications have become an integral part of our lives. ML applications extensively use floating-point computation and involve very large/small numbers; thus, maintaining the numerical stability of such complex computations remains an important challenge. Numerical bugs can lead to system crashes, incorrect output, and wasted computing resources. In this paper, we introduce a novel idea, namely soft assertions (SA), to encode safety/error conditions for the places where numerical instability can occur. A soft assertion is an ML model automatically trained using the dataset obtained during unit testing of unstable functions. Given the values at the unstable function in an ML application, a soft assertion reports how to change these values in order to trigger the instability. We then use the output of soft assertions as signals to effectively mutate inputs to trigger numerical instability in ML applications. In the evaluation, we used the GRIST benchmark, a total of 79 programs, as well as 15 real-world ML applications from GitHub. We compared our tool with 5 state-of-the-art (SOTA) fuzzers. We found all the GRIST bugs and outperformed the baselines. We found 13 numerical bugs in real-world code, one of which had already been confirmed by the GitHub developers. While the baselines mostly found the bugs that report NaN and INF, our tool \tool found numerical bugs with incorrect output. We showed one case where the Tumor Detection Model, trained on Brain MRI images, should have predicted "tumor", but instead, it incorrectly predicted "no tumor" due to the numerical bugs. Our replication package is located at https://figshare.com/s/6528d21ccd28bea94c32.

cs.SE↗

arXiv2024

Closing the Gap: A User Study on the Real-world Usefulness of AI-powered Vulnerability Detection & Repair in the IDE

This paper presents the first empirical study of a vulnerability detection and fix tool with professional software developers on real projects that they own. We implemented DeepVulGuard, an IDE-integrated tool based on state-of-the-art detection and fix models, and show that it has promising performance on benchmarks of historic vulnerability data. DeepVulGuard scans code for vulnerabilities (including identifying the vulnerability type and vulnerable region of code), suggests fixes, provides natural-language explanations for alerts and fixes, leveraging chat interfaces. We recruited 17 professional software developers at Microsoft, observed their usage of the tool on their code, and conducted interviews to assess the tool's usefulness, speed, trust, relevance, and workflow integration. We also gathered detailed qualitative feedback on users' perceptions and their desired features. Study participants scanned a total of 24 projects, 6.9k files, and over 1.7 million lines of source code, and generated 170 alerts and 50 fix suggestions. We find that although state-of-the-art AI-powered detection and fix tools show promise, they are not yet practical for real-world use due to a high rate of false positives and non-applicable fixes. User feedback reveals several actionable pain points, ranging from incomplete context to lack of customization for the user's codebase. Additionally, we explore how AI features, including confidence scores, explanations, and chat interaction, can apply to vulnerability detection and fixing. Based on these insights, we offer practical recommendations for evaluating and deploying AI detection and fix models. Our code and data are available at https://doi.org/10.6084/m9.figshare.26367139.

cs.SE↗

arXiv2024

Testing GPU Numerics: Finding Numerical Differences Between NVIDIA and AMD GPUs

As scientific codes are ported between GPU platforms, continuous testing is required to ensure numerical robustness and identify numerical differences. Compiler-induced numerical differences occur when a program is compiled and run on different GPUs, and the numerical outcomes are different for the same input. We present a study of compiler-induced numerical differences between NVIDIA and AMD GPUs. Our approach uses Varity to generate thousands of short numerical tests in CUDA and HIP, and their inputs; then, we use differential testing to check if the program produced a numerical inconsistency when run on these GPUs. We also use the HIPIFY tool to convert CUDA tests into HIP and check if there are numerical inconsistencies induced by HIPIFY. We generated more than 600,000 tests and found subtle numerical differences that come from (1) math library calls, (2) differences in floating-point precision (FP64 versus FP32), and (3) converting code with HIPIFY.

math.NA↗

arXiv2024

To Err is Machine: Vulnerability Detection Challenges LLM Reasoning

In this paper, we present a challenging code reasoning task: vulnerability detection. Large Language Models (LLMs) have shown promising results in natural-language and math reasoning, but state-of-the-art (SOTA) models reported only 54.5% Balanced Accuracy in our vulnerability detection evaluation, even those models pre-trained on large amounts of source code. Our error analysis on LLM responses shows that the models struggle to reason about the code semantics relevant to identifying vulnerabilities, especially subtle semantic differences caused by small textual changes. We explored prominent models and training settings to understand their effects on vulnerability detection performance -- including better prompts, larger models, more pre-training data, and fine-tuning -- but none led to significant improvements. This raises the question of whether simply scaling training data and model size will allow us to "solve" complex code reasoning tasks like vulnerability detection, or if a fundamental shift in modeling and training techniques is required. We also explored adding domain knowledge to prompts; although it helped certain models understand some code semantics, vulnerability detection requires multi-step reasoning, and these models still failed in steps, such as reasoning about variable relations. Our results suggest that new models, new training methods, or more execution-specific pretraining data may be needed to conquer vulnerability detection. We speculate that auto-regressive pre-training on source code may not effectively extract code semantics, especially on the current pretraining mixtures, in which execution data is scarce. Success on vulnerability detection as a code reasoning task can benefit many areas of software engineering such as debugging, test input generation, and program repair. Our code and data are available at https://doi.org/10.6084/m9.figshare.27368025.

cs.SE↗

arXiv2023

ActiveClean: Generating Line-Level Vulnerability Data via Active Learning

Deep learning vulnerability detection tools are increasing in popularity and have been shown to be effective. These tools rely on large volume of high quality training data, which are very hard to get. Most of the currently available datasets provide function-level labels, reporting whether a function is vulnerable or not vulnerable. However, for a vulnerability detection to be useful, we need to also know the lines that are relevant to the vulnerability. This paper makes efforts towards developing systematic tools and proposes. ActiveClean to generate the large volume of line-level vulnerability data from commits. That is, in addition to function-level labels, it also reports which lines in the function are likely responsible for vulnerability detection. In the past, static analysis has been applied to clean commits to generate line-level data. Our approach based on active learning, which is easy to use and scalable, provide a complementary approach to static analysis. We designed semantic and syntactic properties from commit lines and use them to train the model. We evaluated our approach on both Java and C datasets processing more than 4.3K commits and 119K commit lines. AcitveClean achieved an F1 score between 70-74. Further, we also show that active learning is effective by using just 400 training data to reach F1 score of 70.23. Using ActiveClean, we generate the line-level labels for the entire FFMpeg project in the Devign dataset, including 5K functions, and also detected incorrect function-level labels. We demonstrated that using our cleaned data, LineVul, a SOTA line-level vulnerability detection tool, detected 70 more vulnerable lines and 18 more vulnerable functions, and improved Top 10 accuracy from 66% to 73%.

cs.SE↗

arXiv2023

Do Language Models Learn Semantics of Code? A Case Study in Vulnerability Detection

Recently, pretrained language models have shown state-of-the-art performance on the vulnerability detection task. These models are pretrained on a large corpus of source code, then fine-tuned on a smaller supervised vulnerability dataset. Due to the different training objectives and the performance of the models, it is interesting to consider whether the models have learned the semantics of code relevant to vulnerability detection, namely bug semantics, and if so, how the alignment to bug semantics relates to model performance. In this paper, we analyze the models using three distinct methods: interpretability tools, attention analysis, and interaction matrix analysis. We compare the models' influential feature sets with the bug semantic features which define the causes of bugs, including buggy paths and Potentially Vulnerable Statements (PVS). We find that (1) better-performing models also aligned better with PVS, (2) the models failed to align strongly to PVS, and (3) the models failed to align at all to buggy paths. Based on our analysis, we developed two annotation methods which highlight the bug semantics inside the model's inputs. We evaluated our approach on four distinct transformer models and four vulnerability datasets and found that our annotations improved the models' performance in the majority of settings - 11 out of 16, with up to 9.57 points improvement in F1 score compared to conventional fine-tuning. We further found that with our annotations, the models aligned up to 232% better to potentially vulnerable statements. Our findings indicate that it is helpful to provide the model with information of the bug semantics, that the model can attend to it, and motivate future work in learning more complex path-based bug semantics. Our code and data are available at https://figshare.com/s/4a16a528d6874aad51a0.

cs.LG↗

arXiv2023

Towards Causal Deep Learning for Vulnerability Detection

Deep learning vulnerability detection has shown promising results in recent years. However, an important challenge that still blocks it from being very useful in practice is that the model is not robust under perturbation and it cannot generalize well over the out-of-distribution (OOD) data, e.g., applying a trained model to unseen projects in real world. We hypothesize that this is because the model learned non-robust features, e.g., variable names, that have spurious correlations with labels. When the perturbed and OOD datasets no longer have the same spurious features, the model prediction fails. To address the challenge, in this paper, we introduced causality into deep learning vulnerability detection. Our approach CausalVul consists of two phases. First, we designed novel perturbations to discover spurious features that the model may use to make predictions. Second, we applied the causal learning algorithms, specifically, do-calculus, on top of existing deep learning models to systematically remove the use of spurious features and thus promote causal based prediction. Our results show that CausalVul consistently improved the model accuracy, robustness and OOD performance for all the state-of-the-art models and datasets we experimented. To the best of our knowledge, this is the first work that introduces do calculus based causal learning to software engineering models and shows it's indeed useful for improving the model accuracy, robustness and generalization. Our replication package is located at https://figshare.com/s/0ffda320dcb96c249ef2.

cs.SE↗

arXiv2023

MixQuant: Mixed Precision Quantization with a Bit-width Optimization Search

Quantization is a technique for creating efficient Deep Neural Networks (DNNs), which involves performing computations and storing tensors at lower bit-widths than f32 floating point precision. Quantization reduces model size and inference latency, and therefore allows for DNNs to be deployed on platforms with constrained computational resources and real-time systems. However, quantization can lead to numerical instability caused by roundoff error which leads to inaccurate computations and therefore, a decrease in quantized model accuracy. Similarly to prior works, which have shown that both biases and activations are more sensitive to quantization and are best kept in full precision or quantized with higher bit-widths, we show that some weights are more sensitive than others which should be reflected on their quantization bit-width. To that end we propose MixQuant, a search algorithm that finds the optimal custom quantization bit-width for each layer weight based on roundoff error and can be combined with any quantization method as a form of pre-processing optimization. We show that combining MixQuant with BRECQ, a state-of-the-art quantization method, yields better quantized model accuracy than BRECQ alone. Additionally, we combine MixQuant with vanilla asymmetric quantization to show that MixQuant has the potential to optimize the performance of any quantization technique.

cs.LG↗

arXiv2023

Reproducing Failures in Fault Signatures

Software often fails in the field, however reproducing and debugging field failures is very challenging: the failure-inducing input may be missing, and the program setup can be complicated and hard to reproduce by the developers. In this paper, we propose to generate fault signatures from the failure locations and the original source code to reproduce the faults in small executable programs. We say that a fault signature reproduces the fault in the original program if the two failed in the same location, triggered the same error conditions after executing the same selective sequences of failure-inducing statements. A fault signature aims to contain only sufficient statements that can reproduce the faults. That way, it provides some context to inform how a fault is developed and also avoids unnecessary complexity and setups that may block fault diagnosis. To compute fault signatures from the failures, we applied a path-sensitive static analysis tool to generate a path that leads to the fault, and then applied an existing syntactic patching tool to convert the path into an executable program. Our evaluation on real-world bugs from Corebench, BugBench, and Manybugs shows that fault signatures can reproduce the fault for the original programs. Because fault signatures are less complex, automatic test input generation tools generated failure-inducing inputs that could not be generated by using the entire programs. Some failure-inducing inputs can be directly transferred to the original programs. Our experimental data are publicly available at https://doi.org/10.5281/zenodo.5430155.

cs.SE↗

arXiv2023

An Effective Data-Driven Approach for Localizing Deep Learning Faults

Deep Learning (DL) applications are being used to solve problems in critical domains (e.g., autonomous driving or medical diagnosis systems). Thus, developers need to debug their systems to ensure that the expected behavior is delivered. However, it is hard and expensive to debug DNNs. When the failure symptoms or unsatisfied accuracies are reported after training, we lose the traceability as to which part of the DNN program is responsible for the failure. Even worse, sometimes, a deep learning program has different types of bugs. To address the challenges of debugging DNN models, we propose a novel data-driven approach that leverages model features to learn problem patterns. Our approach extracts these features, which represent semantic information of faults during DNN training. Our technique uses these features as a training dataset to learn and infer DNN fault patterns. Also, our methodology automatically links bug symptoms to their root causes, without the need for manually crafted mappings, so that developers can take the necessary steps to fix faults. We evaluate our approach using real-world and mutated models. Our results demonstrate that our technique can effectively detect and diagnose different bug types. Finally, our technique achieved better accuracy, precision, and recall than prior work for mutated models. Also, our approach achieved comparable results for real-world models in terms of accuracy and performance to the state-of-the-art.

cs.SE↗

arXiv2023

TRACED: Execution-aware Pre-training for Source Code

Most existing pre-trained language models for source code focus on learning the static code text, typically augmented with static code structures (abstract syntax tree, dependency graphs, etc.). However, program semantics will not be fully exposed before the real execution. Without an understanding of the program execution, statically pre-trained models fail to comprehensively capture the dynamic code properties, such as the branch coverage and the runtime variable values, and they are consequently less effective at code understanding tasks, such as retrieving semantic clones and detecting software vulnerabilities. To close the gap between the static nature of language models and the dynamic characteristics of programs, we introduce TRACED, an execution-aware pre-training strategy for source code. Specifically, we pre-train code language models with a combination of source code, executable inputs, and corresponding execution traces. Our goal is to teach code models the complicated execution logic during the pre-training, enabling the model to statically estimate the dynamic code properties without repeatedly executing code during task-specific fine-tuning. To illustrate the effectiveness of our proposed approach, we fine-tune and evaluate TRACED on three downstream tasks: static execution estimation, clone retrieval, and vulnerability detection. The empirical results show that TRACED relatively improves the statically pre-trained code models by 12.4% for complete execution path prediction and by 25.2% for runtime variable value predictions. TRACED also significantly outperforms statically pre-trained models in clone retrieval and vulnerability detection across four public benchmarks.

cs.SE↗

arXiv2023

Beryllium: Neural Search for Algorithm Implementations

In this paper, we explore the feasibility of finding algorithm implementations from code. Successfully matching code and algorithms can help understand unknown code, provide reference implementations, and automatically collect data for learning-based program synthesis. To achieve the goal, we designed a new language named p-language to specify the algorithms and a static analyzer for the p-language to automatically extract control flow, math, and natural language information from the algorithm descriptions. We embedded the output of p-language (p-code) and source code in a common vector space using self-supervised machine learning methods to match algorithm with code without any manual annotation. We developed a tool named Beryllium. It takes pseudo code as a query and returns a list of ranked code snippets that likely match the algorithm query. Our evaluation on Stony Brook Algorithm Repository and popular GitHub projects show that Beryllium significantly outperformed the state-of-the-art code search tools in both C and Java. Specifically, for 98.5%, 93.8%, and 66.2% queries, we found the algorithm implementations in the top 25, 10, and 1 ranked list, respectively. Given 87 algorithm queries, we found implementations for 74 algorithms in the GitHub projects where we did not know the algorithms before.

cs.SE↗

arXiv2023

A Study of Static Warning Cascading Tools (Experience Paper)

Static analysis is widely used for software assurance. However, static analysis tools can report an overwhelming number of warnings, many of which are false positives. Applying static analysis to a new version, a large number of warnings can be only relevant to the old version. Inspecting these warnings is a waste of time and can prevent developers from finding the new bugs in the new version. In this paper, we report the challenges of cascading warnings generated from two versions of programs. We investigated program differencing tools and extend them to perform warning cascading automatically. Specifically, we used textual based diff tool, namely SCALe, abstract syntax tree (AST) based diff tool, namely GumTree, and control flow graph (CFG) based diff tool, namely Hydrogen. We reported our experience of applying these tools and hopefully our findings can provide developers understandings of pros and cons of each approach. In our evaluation, we used 96 pairs of benchmark programs for which we know ground-truth bugs and fixes as well as 12 pairs of real-world open-source projects. Our tools and data are available at https: //github.com/WarningCas/WarningCascading_Data.

cs.SE↗