arXiv subjects

Yanru Xiao

Publications and source records attributed to Yanru Xiao.

5 recordsLinked to original sources

arXiv2026

Follow the Policy, Fail the Benchmark: Auditing and Revising Policy-Based Decision Benchmarks

Policy-based decision benchmarks evaluate whether systems apply written rules to financial and operational cases. Yet a rule-following response can be marked wrong when the benchmark misstates the required decision. We audit nine public policy-based decision benchmarks through source tracing, program execution, and controlled checks. We find unsupported targets, reference errors that survive replay or propagate as solved steps, and evaluators that reject acceptable answers or credit incomplete decisions. To measure the scoring impact, we rescore unchanged model responses. On the 96 LegalBench SARA numeric cases, correcting the scorer raises acceptance by more than 80 percentage points for each of three models, without changing references or responses. In RuleArena Tax, restoring a required zero floor changes 27 of 300 references and narrows an observed model gap. A controlled federal-exemption task with explicit ownership and eligibility assumptions also shows that errors in supplied intermediate answers can affect final decisions. We present BAR (Benchmark Audit and Revision), a versioned collection of audit evidence and revisions for reevaluation.

cs.CL↗

arXiv2026

Cognitive Bias Signatures in Search-Equipped LLM Agents: A Stage-Resolved Analysis

Search can help LLM agents correct mistakes, but their starting conditions may shape which evidence they find and use. Drawing on cognitive-bias paradigms, we develop a controlled search harness to examine result order, records of prior work, and initial hypotheses while holding each question and document collection fixed. Matched controls distinguish order effects from differences in document composition and the effects of failed work from context length. Across 10 models and 1,000 synthetic search worlds, we track queries, retrieved documents, document openings, and final answers. Our comparisons suggest that final answers are more sensitive to initial evidence order during active search than when all evidence is supplied in full. Dead-end histories can reduce strategy switching relative to matched neutral histories, indicating an effect beyond context length. Under wrong hypotheses, stage measurements distinguish reduced retrieval or opening from errors that persist after counter-evidence is opened. Exploratory interventions show that opening counter-evidence more often can coincide with lower accuracy, while accuracy can improve with little change in opening.

cs.CL↗

arXiv2026

Vision-TTT: Efficient and Expressive Visual Representation Learning with Test-Time Training

Learning efficient and expressive visual representation has long been the pursuit of computer vision research. While Vision Transformers (ViTs) gradually replace traditional Convolutional Neural Networks (CNNs) as more scalable vision learners, their applications are plagued by the quadratic complexity of the self-attention mechanism. To address the challenge, we introduce a new linear-time sequence modeling method Test-Time Training (TTT) into vision and propose Vision-TTT, which treats visual sequences as datasets and compresses the visual token sequences in a novel self-supervised learning manner. By incorporating the dual-dataset strategy and Conv2d-based dataset preprocessing, Vision-TTT effectively extends vanilla TTT to model 2D visual correlations with global receptive fields. Extensive experiments show that \texttt{Vittt-T/S/B} achieve $77.7\%,81.8\%,82.7\%$ Top-1 accuracy on ImageNet classification and also greatly outperform their counterparts on downstream tasks. At $1280\times1280$ resolution, \texttt{Vittt-T} reduces FLOPs by $79.4\%$ and runs $4.72\times$ faster with $88.9\%$ less memory than DeiT-T. These results demonstrate the expressiveness and efficiency of Vision-TTT as a strong candidate for the next-generation generic visual backbone.

cs.CV↗

arXiv2023

Unsupervised Multi-Criteria Adversarial Detection in Deep Image Retrieval

The vulnerability in the algorithm supply chain of deep learning has imposed new challenges to image retrieval systems in the downstream. Among a variety of techniques, deep hashing is gaining popularity. As it inherits the algorithmic backend from deep learning, a handful of attacks are recently proposed to disrupt normal image retrieval. Unfortunately, the defense strategies in softmax classification are not readily available to be applied in the image retrieval domain. In this paper, we propose an efficient and unsupervised scheme to identify unique adversarial behaviors in the hamming space. In particular, we design three criteria from the perspectives of hamming distance, quantization loss and denoising to defend against both untargeted and targeted attacks, which collectively limit the adversarial space. The extensive experiments on four datasets demonstrate 2-23% improvements of detection rates with minimum computational overhead for real-time image queries.

cs.CV↗

arXiv2020

A Framework for Behavioral Biometric Authentication using Deep Metric Learning on Mobile Devices

Mobile authentication using behavioral biometrics has been an active area of research. Existing research relies on building machine learning classifiers to recognize an individual's unique patterns. However, these classifiers are not powerful enough to learn the discriminative features. When implemented on the mobile devices, they face new challenges from the behavioral dynamics, data privacy and side-channel leaks. To address these challenges, we present a new framework to incorporate training on battery-powered mobile devices, so private data never leaves the device and training can be flexibly scheduled to adapt the behavioral patterns at runtime. We re-formulate the classification problem into deep metric learning to improve the discriminative power and design an effective countermeasure to thwart side-channel leaks by embedding a noise signature in the sensing signals without sacrificing too much usability. The experiments demonstrate authentication accuracy over 95% on three public datasets, a sheer 15% gain from multi-class classification with less data and robustness against brute-force and side-channel attacks with 99% and 90% success, respectively. We show the feasibility of training with mobile CPUs, where training 100 epochs takes less than 10 mins and can be boosted 3-5 times with feature transfer. Finally, we profile memory, energy and computational overhead. Our results indicate that training consumes lower energy than watching videos and slightly higher energy than playing games.

cs.LG↗