arXiv subjects

Jason Zhang

Publications and source records attributed to Jason Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Approximate Algorithms for Chamfer Distance Under Translation

Given two sets of points A and B, $|A| = m$, $|B| = n$, the Chamfer distance from $A$ to $B$ is defined as $\operatorname{CD}(A,B) = \sum_{a\in A} \min_{b\in B} d(a,b)$, where $d$ is a distance metric. Chamfer distance is a popular measure of dissimilarity between two sets of points that has seen increasing usage in computer vision and information retrieval as a substitute for the more computationally demanding Earth Mover's distance. We propose a new problem, Chamfer distance under translation, defined as $\operatorname{CDuT}(A,B) :=\min_{t\in \mathbb{R}^d} \operatorname{CD}(A+t,B)$, where $A+t$ denotes the translation of every point in $A$ by $t$. Chamfer distance under translation is valuable in cases where translations capture aspects of the data unlikely to be relevant for dissimilarity, such as temporal, spatial, or other semantic information. For Chamfer distance under translation, we provide four algorithms: (1) an exact quadratic time algorithm in one dimension, (2) a near quadratic time ($2+\varepsilon$)-approximation algorithm in higher dimensions, (3) a $(1+\varepsilon)$-approximation algorithm with running time $\mathcal{O}(mn^2\varepsilon^{-(d+1)})$, and (4) a near-quadratic time $(1+\varepsilon)$-approximation algorithm for answering the decision version of $\operatorname{CDuT}$ given a separation assumption on $B$. We additionally explore the fine-grained complexity of $\operatorname{CDuT}$.

cs.DS↗

arXiv2026

Evaluating AI Meeting Summaries with a Reusable Cross-Domain Pipeline

Industrial teams often deploy large language model features before stable regression or model selection evaluation exists. We present a reusable evaluation system for AI meeting summaries that combines structured ground-truth (GT) construction, fixed candidate generation, claim-grounded scoring, persisted reporting, and a privacy-bounded online monitoring and nomination interface. The online evidence is not itself a benchmark: privacy-safe aggregate exports show active monitoring, hard regime detection, and directional movement without exposing customer data. We benchmark the offline path on 114 meetings across city_council, private_data, and whitehouse_press_briefings, yielding 340 completed meeting-model pairs and 680 judge runs for gpt-4.1-mini, gpt-5-mini, and gpt-5.1. Under this fixed protocol, accuracy differences are not statistically significant under Holm correction (corrected p-values 0.053-0.448), although gpt-4.1-mini has the highest mean accuracy (0.583); the significant separation is on retention, where gpt-5.1 leads on completeness (0.886) and coverage (0.942). Typed slices isolate whitehouse_press_briefings as an accuracy-hard regime, and a later focused rerun over gpt-4.1, gpt-5-mini, and gpt-5.4 reuses the same stack under the same judges and metrics. This extended preprint keeps those core results aligned with the formal submission while adding a more detailed repository-level account of cross-domain reuse from the companion AI-search paper and an additional typed DeepEval contrastive analysis. Model naming note. Running text uses canonical model names on first introduction. Tables, filenames, and artifact IDs retain compact report labels for consistency with the packaged benchmark outputs. Table A maps the two conventions and is repeated in Section 4.3 where candidate-generation settings are defined.

cs.AI↗

arXiv2026

One Panel Does Not Fit All: Case-Adaptive Multi-Agent Deliberation for Clinical Prediction

Large language models applied to clinical prediction exhibit case-level heterogeneity: simple cases yield consistent outputs, while complex cases produce divergent predictions under minor prompt changes. Existing single-agent strategies sample from one role-conditioned distribution, and multi-agent frameworks use fixed roles with flat majority voting, discarding the diagnostic signal in disagreement. We propose CAMP (Case-Adaptive Multi-agent Panel), where an attending-physician agent dynamically assembles a specialist panel tailored to each case's diagnostic uncertainty. Each specialist evaluates candidates via three-valued voting (KEEP/REFUSE/NEUTRAL), enabling principled abstention outside one's expertise. A hybrid router directs each diagnosis through strong consensus, fallback to the attending physician's judgment, or evidence-based arbitration that weighs argument quality over vote counts. On diagnostic prediction and brief hospital course generation from MIMIC-IV across four LLM backbones, CAMP consistently outperforms strong baselines while consuming fewer tokens than most competing multi-agent methods, with voting records and arbitration traces offering transparent decision audits.

cs.AI↗

arXiv2025

VLIC: Vision-Language Models As Perceptual Judges for Human-Aligned Image Compression

Evaluations of image compression performance which include human preferences have generally found that naive distortion functions such as MSE are insufficiently aligned to human perception. In order to align compression models to human perception, prior work has employed differentiable perceptual losses consisting of neural networks calibrated on large-scale datasets of human psycho-visual judgments. We show that, surprisingly, state-of-the-art vision-language models (VLMs) can replicate binary human two-alternative forced choice (2AFC) judgments zero-shot when asked to reason about the differences between pairs of images. Motivated to exploit the powerful zero-shot visual reasoning capabilities of VLMs, we propose Vision-Language Models for Image Compression (VLIC), a diffusion-based image compression system designed to be post-trained with binary VLM judgments. VLIC leverages existing techniques for diffusion model post-training with preferences, rather than distilling the VLM judgments into a separate perceptual loss network. We show that calibrating this system on VLM judgments produces competitive or state-of-the-art performance on human-aligned visual compression depending on the dataset, according to perceptual metrics and large-scale user studies. We additionally conduct an extensive analysis of the VLM-based reward design and training procedure and share important insights. More visuals are available at https://kylesargent.github.io/vlic

cs.CV↗

arXiv2025

Is Passive Expertise-Based Personalization Enough? A Case Study in AI-Assisted Test-Taking

Novice and expert users have different systematic preferences in task-oriented dialogues. However, whether catering to these preferences actually improves user experience and task performance remains understudied. To investigate the effects of expertise-based personalization, we first built a version of an enterprise AI assistant with passive personalization. We then conducted a user study where participants completed timed exams, aided by the two versions of the AI assistant. Preliminary results indicate that passive personalization helps reduce task load and improve assistant perception, but reveal task-specific limitations that can be addressed through providing more user agency. These findings underscore the importance of combining active and passive personalization to optimize user experience and effectiveness in enterprise task-oriented environments.

cs.HC↗

arXiv2025

AutoLibra: Agent Metric Induction from Open-Ended Human Feedback

Agents are predominantly evaluated and optimized via task success metrics, which are coarse, rely on manual design from experts, and fail to reward intermediate emergent behaviors. We propose **AutoLibra**, a framework for agent evaluation, that transforms open-ended human feedback *e.g.* "If you find that the button is disabled, don't click it again", or "This agent has too much autonomy to decide what to do on its own" into metrics for evaluating fine-grained behaviors in agent trajectories. AutoLibra accomplishes this by grounding feedback to an agent's behavior, clustering similar positive and negative behaviors, and creating concrete metrics with clear definitions and concrete examples, which can be used for prompting LLM-as-a-Judge as evaluators. We further propose two meta metrics to evaluate the alignment of a set of (induced) metrics with open feedback: "coverage" and "redundancy". Through optimizing these meta-metrics, we experimentally demonstrate AutoLibra's ability to induce more concrete agent evaluation metrics than the ones proposed in previous agent evaluation benchmarks and discover new metrics to analyze agents. We also present two applications of AutoLibra in agent improvement: First, we show that AutoLibra serve human prompt engineers for diagonalize agent failures and improve prompts iterative. Moreover, we find that AutoLibra can induce metrics for automatic optimization for agents, which makes agents improve through self-regulation. Our results suggest that AutoLibra is a powerful task-agnostic tool for evaluating and improving language agents.

cs.AI↗

arXiv2025

The Structural Safety Generalization Problem

LLM jailbreaks are a widespread safety challenge. Given this problem has not yet been tractable, we suggest targeting a key failure mechanism: the failure of safety to generalize across semantically equivalent inputs. We further focus the target by requiring desirable tractability properties of attacks to study: explainability, transferability between models, and transferability between goals. We perform red-teaming within this framework by uncovering new vulnerabilities to multi-turn, multi-image, and translation-based attacks. These attacks are semantically equivalent by our design to their single-turn, single-image, or untranslated counterparts, enabling systematic comparisons; we show that the different structures yield different safety outcomes. We then demonstrate the potential for this framework to enable new defenses by proposing a Structure Rewriting Guardrail, which converts an input to a structure more conducive to safety assessment. This guardrail significantly improves refusal of harmful inputs, without over-refusing benign ones. Thus, by framing this intermediate challenge - more tractable than universal defenses but essential for long-term safety - we highlight a critical milestone for AI safety research.

cs.CR↗

arXiv2024

Uncovering Latent Chain of Thought Vectors in Language Models

In this work, we examine how targeted perturbations in the activation space of Language Models (LMs) can encode complex reasoning patterns. We inject steering vectors, derived from LM activations, into LMs during inference time and study whether these vectors can induce Chain-of-Thought (CoT) reasoning in LMs without the need for natural language prompting. We demonstrate this approach on Llama3 8B Instruct and Mistral 7B v0.2 Instruct and show that activation-space interventions achieve competitive, if not superior, performance compared to traditional CoT prompting across multiple reasoning benchmarks, including GSM8k, MMLU, AGI Eval, and ARC AI2. These findings suggest that neural network activations can encode reasoning patterns, offering a new application of activation space manipulation as a tool for tuning model behavior.

cs.CL↗

arXiv2024

Emerging Vulnerabilities in Frontier Models: Multi-Turn Jailbreak Attacks

Large language models (LLMs) are improving at an exceptional rate. However, these models are still susceptible to jailbreak attacks, which are becoming increasingly dangerous as models become increasingly powerful. In this work, we introduce a dataset of jailbreaks where each example can be input in both a single or a multi-turn format. We show that while equivalent in content, they are not equivalent in jailbreak success: defending against one structure does not guarantee defense against the other. Similarly, LLM-based filter guardrails also perform differently depending on not just the input content but the input structure. Thus, vulnerabilities of frontier models should be studied in both single and multi-turn settings; this dataset provides a tool to do so.

cs.CR↗

arXiv2024

Tight Lower Bounds for Directed Cut Sparsification and Distributed Min-Cut

In this paper, we consider two fundamental cut approximation problems on large graphs. We prove new lower bounds for both problems that are optimal up to logarithmic factors. The first problem is to approximate cuts in balanced directed graphs. In this problem, the goal is to build a data structure that $(1 \pm \epsilon)$-approximates cut values in graphs with $n$ vertices. For arbitrary directed graphs, such a data structure requires $\Omega(n^2)$ bits even for constant $\epsilon$. To circumvent this, recent works study $\beta$-balanced graphs, meaning that for every directed cut, the total weight of edges in one direction is at most $\beta$ times that in the other direction. We consider two models: the {\em for-each} model, where the goal is to approximate each cut with constant probability, and the {\em for-all} model, where all cuts must be preserved simultaneously. We improve the previous $\Omega(n \sqrt{\beta/\epsilon})$ lower bound to $\tilde{\Omega}(n \sqrt{\beta}/\epsilon)$ in the for-each model, and we improve the previous $\Omega(n \beta/\epsilon)$ lower bound to $\Omega(n \beta/\epsilon^2)$ in the for-all model. This resolves the main open questions of (Cen et al., ICALP, 2021). The second problem is to approximate the global minimum cut in a local query model, where we can only access the graph via degree, edge, and adjacency queries. We improve the previous $\Omega\bigl(\frac{m}{k}\bigr)$ query complexity lower bound to $\Omega\bigl(\min\{m, \frac{m}{\epsilon^2 k}\}\bigr)$ for this problem, where $m$ is the number of edges, $k$ is the size of the minimum cut, and we seek a $(1+\epsilon)$-approximation. In addition, we show that existing upper bounds with slight modifications match our lower bound up to logarithmic factors.

cs.DS↗

arXiv2023

Accelerating Exploration with Unlabeled Prior Data

Learning to solve tasks from a sparse reward signal is a major challenge for standard reinforcement learning (RL) algorithms. However, in the real world, agents rarely need to solve sparse reward tasks entirely from scratch. More often, we might possess prior experience to draw on that provides considerable guidance about which actions and outcomes are possible in the world, which we can use to explore more effectively for new tasks. In this work, we study how prior data without reward labels may be used to guide and accelerate exploration for an agent solving a new sparse reward task. We propose a simple approach that learns a reward model from online experience, labels the unlabeled prior data with optimistic rewards, and then uses it concurrently alongside the online data for downstream policy and critic optimization. This general formula leads to rapid exploration in several challenging sparse-reward domains where tabula rasa exploration is insufficient, including the AntMaze domain, Adroit hand manipulation domain, and a visual simulated robotic manipulation domain. Our results highlight the ease of incorporating unlabeled prior data into existing online RL algorithms, and the (perhaps surprising) effectiveness of doing so.

cs.LG↗

arXiv2023

Empirical Evaluation of the Segment Anything Model (SAM) for Brain Tumor Segmentation

Brain tumor segmentation presents a formidable challenge in the field of Medical Image Segmentation. While deep-learning models have been useful, human expert segmentation remains the most accurate method. The recently released Segment Anything Model (SAM) has opened up the opportunity to apply foundation models to this difficult task. However, SAM was primarily trained on diverse natural images. This makes applying SAM to biomedical segmentation, such as brain tumors with less defined boundaries, challenging. In this paper, we enhanced SAM's mask decoder using transfer learning with the Decathlon brain tumor dataset. We developed three methods to encapsulate the four-dimensional data into three dimensions for SAM. An on-the-fly data augmentation approach has been used with a combination of rotations and elastic deformations to increase the size of the training dataset. Two key metrics: the Dice Similarity Coefficient (DSC) and the Hausdorff Distance 95th Percentile (HD95), have been applied to assess the performance of our segmentation models. These metrics provided valuable insights into the quality of the segmentation results. In our evaluation, we compared this improved model to two benchmarks: the pretrained SAM and the widely used model, nnUNetv2. We find that the improved SAM shows considerable improvement over the pretrained SAM, while nnUNetv2 outperformed the improved SAM in terms of overall segmentation accuracy. Nevertheless, the improved SAM demonstrated slightly more consistent results than nnUNetv2, especially on challenging cases that can lead to larger Hausdorff distances. In the future, more advanced techniques can be applied in order to further improve the performance of SAM on brain tumor segmentation.

eess.IV↗

arXiv2023

A 5' UTR Language Model for Decoding Untranslated Regions of mRNA and Function Predictions

The 5' UTR, a regulatory region at the beginning of an mRNA molecule, plays a crucial role in regulating the translation process and impacts the protein expression level. Language models have showcased their effectiveness in decoding the functions of protein and genome sequences. Here, we introduced a language model for 5' UTR, which we refer to as the UTR-LM. The UTR-LM is pre-trained on endogenous 5' UTRs from multiple species and is further augmented with supervised information including secondary structure and minimum free energy. We fine-tuned the UTR-LM in a variety of downstream tasks. The model outperformed the best-known benchmark by up to 42% for predicting the Mean Ribosome Loading, and by up to 60% for predicting the Translation Efficiency and the mRNA Expression Level. The model also applies to identifying unannotated Internal Ribosome Entry Sites within the untranslated region and improves the AUPR from 0.37 to 0.52 compared to the best baseline. Further, we designed a library of 211 novel 5' UTRs with high predicted values of translation efficiency and evaluated them via a wet-lab assay. Experiment results confirmed that our top designs achieved a 32.5% increase in protein production level relative to well-established 5' UTR optimized for therapeutics.

cs.LG↗

arXiv2023

Domain Knowledge Distillation from Large Language Model: An Empirical Study in the Autonomous Driving Domain

Engineering knowledge-based (or expert) systems require extensive manual effort and domain knowledge. As Large Language Models (LLMs) are trained using an enormous amount of cross-domain knowledge, it becomes possible to automate such engineering processes. This paper presents an empirical automation and semi-automation framework for domain knowledge distillation using prompt engineering and the LLM ChatGPT. We assess the framework empirically in the autonomous driving domain and present our key observations. In our implementation, we construct the domain knowledge ontology by "chatting" with ChatGPT. The key finding is that while fully automated domain ontology construction is possible, human supervision and early intervention typically improve efficiency and output quality as they lessen the effects of response randomness and the butterfly effect. We, therefore, also develop a web-based distillation assistant enabling supervision and flexible intervention at runtime. We hope our findings and tools could inspire future research toward revolutionizing the engineering of knowledge-based systems across application domains.

cs.CL↗

arXiv2023

The Linked Data Benchmark Council (LDBC): Driving competition and collaboration in the graph data management space

Graph data management is instrumental for several use cases such as recommendation, root cause analysis, financial fraud detection, and enterprise knowledge representation. Efficiently supporting these use cases yields a number of unique requirements, including the need for a concise query language and graph-aware query optimization techniques. The goal of the Linked Data Benchmark Council (LDBC) is to design a set of standard benchmarks that capture representative categories of graph data management problems, making the performance of systems comparable and facilitating competition among vendors. LDBC also conducts research on graph schemas and graph query languages. This paper introduces the LDBC organization and its work over the last decade.

cs.DB↗

arXiv2022

Using Deep Mixture-of-Experts to Detect Word Meaning Shift for TempoWiC

This paper mainly describes the dma submission to the TempoWiC task, which achieves a macro-F1 score of 77.05% and attains the first place in this task. We first explore the impact of different pre-trained language models. Then we adopt data cleaning, data augmentation, and adversarial training strategies to enhance the model generalization and robustness. For further improvement, we integrate POS information and word semantic representation using a Mixture-of-Experts (MoE) approach. The experimental results show that MoE can overcome the feature overuse issue and combine the context, POS, and word semantic features well. Additionally, we use a model ensemble method for the final prediction, which has been proven effective by many research works.

cs.CL↗

arXiv2020

Towards Real-time Drowsiness Detection for Elderly Care

The primary focus of this paper is to produce a proof of concept for extracting drowsiness information from videos to help elderly living on their own. To quantify yawning, eyelid and head movement over time, we extracted 3000 images from captured videos for training and testing of deep learning models integrated with OpenCV library. The achieved classification accuracy for eyelid and mouth open/close status were between 94.3%-97.2%. Visual inspection of head movement from videos with generated 3D coordinate overlays, indicated clear spatiotemporal patterns in collected data (yaw, roll and pitch). Extraction methodology of the drowsiness information as timeseries is applicable to other contexts including support for prior work in privacy-preserving augmented coaching, sport rehabilitation, and integration with big data platform in healthcare.

cs.CV↗

arXiv2020

Tethered capsule en face optical coherence tomography for imaging Barrett's esophagus in unsedated patients

Detection of Barrett's esophagus (BE) at points of care outside the endoscopy suite may improve screening access and reduce esophageal adenocarcinoma mortality. Tethered capsule optical coherence tomography (OCT) can volumetrically image esophageal mucosa and detect BE in unsedated patients. We investigated ultrahigh-speed tethered capsule, swept-source OCT (SS-OCT) in unsedated patients, improved device design, developed procedural techniques, measured how capsule contact and longitudinal pullback non-uniformity affect coverage, and assessed patient toleration. OCT was performed in 16 patients prior to endoscopic surveillance/treatment. Unsedated patients swallowed the capsule with small sips of water and the esophagus imaged by retracting the tether. Ultrahigh-speed SS-OCT at 1,000,000 A-scans/second imaged ~40cm$^2$ esophageal areas in 10 seconds with 30um transverse and 8um axial resolution. Nine patients had non-dysplastic BE, 3 had ablative treatment-naive neoplasia, and 4 had prior ablation for history of dysplasia. The capsule procedure was well tolerated. Ultrahigh-speed SS-OCT enabled the generation of cross-sectional and sub-surface en face images. BE could be rapidly identified in cross-sectional and en face images, while assessment of the gastro-esophageal junction (GEJ) required subsurface en face views. Capsule esophageal contact was worse in long-segment BE and sliding hiatal hernia, and longitudinal image coverage was worse in short-segment BE. OCT candidate features of dysplasia are reported. Ultrahigh-speed tethered capsule SS-OCT enabled en face and cross-sectional imaging of mucosal features over wide areas. Device and procedure optimization led to improved imaging performance. Areas of BE could be readily identified, but limited capsule contact and longitudinal image coverage can yield sampling errors.

physics.med-ph↗