arXiv subjects

Ali Ansari

Publications and source records attributed to Ali Ansari.

10 recordsLinked to original sources

arXiv2026

ERUnderstand: Evaluating Vision-Language Models on Structured ER Diagrams

Entity-Relationship Diagrams (ERDs) are central to conceptual database design, yet they are typically available only as rendered images rather than machine-readable schemas, limiting AI-assisted database engineering. We introduce ERUnderstand, the first large-scale benchmark for structured understanding of ER diagrams, comprising 2,960 diagrams collected from curated educational sources, real-world schemas, and synthetically generated examples spanning diverse domains, notations, complexity levels, and Extended Entity-Relationship (EER) constructs. Each diagram is paired with a standardized machine-readable representation for fine-grained evaluation of schema elements. Evaluating state-of-the-art Vision-Language Models (VLMs), we find that while common ERD elements are recovered reliably (F1 > 0.74), performance drops sharply on weak entities (as low as 0.28 F1), multivalued attributes (0.14 F1), and N-ary relationships (0.07 F1). Reasoning-augmented models improve overall performance by 15-25% but remain sensitive to linguistic priors and increasing diagram complexity. ERUnderstand provides a standardized benchmark for evaluating multimodal understanding of conceptual database schemas. The benchmark, dataset, evaluation toolkit, and generation code are publicly available at https://github.com/salinaria/ERUnderstand.

cs.AI↗

arXiv2026

No Last Mile: A Theory of the Human Data Market

The standard framing treats structured human-data work as transitional, a bridge between today's imperfect models and a future state where automation is complete. We challenge this view by modeling structured human data as a persistent production input: evaluation, rubric-based judgment, auditing, exception handling, and continual updates that convert raw model capability into dependable, deployable performance. These activities accumulate into a reusable AI capability stock that raises productivity by improving reliability on existing tasks and by expanding the frontier of task families for which AI can be used at high confidence. Crucially, this capability stock depreciates as tasks and contexts drift, standards evolve, and new edge cases emerge. In a tractable baseline model, an interior steady state implies a closed-form, strictly positive long-run labor share devoted to structured human-data work whenever depreciation is positive, a "no last mile" result in which maintenance demand persists even as models improve. We then microfound aggregate capability with a portfolio of task families featuring diminishing returns, frontier entry, and complementarity, generating reallocation toward low-maturity and bottleneck families and a Roy-style mechanism for within-structured wage dispersion. Finally, we map model objects to observable proxies using standard data layers, and provide a conservative calibration suggesting a 5-7% steady-state structured labor share in the long run.

econ.GN↗

arXiv2025

Evaluating Speech-to-Text x LLM x Text-to-Speech Combinations for AI Interview Systems

Voice-based conversational AI systems increasingly rely on cascaded architectures that combine speech-to-text (STT), large language models (LLMs), and text-to-speech (TTS) components. We present a large-scale empirical comparison of STT x LLM x TTS stacks using data sampled from over 300,000 AI-conducted job interviews. We used an LLM-as-a-Judge automated evaluation framework to assess conversational quality, technical accuracy, and skill assessment capabilities. Our analysis of five production configurations reveals that a stack combining Google's STT, GPT-4.1, and Cartesia's TTS outperforms alternatives in both objective quality metrics and user satisfaction scores. Surprisingly, we find that objective quality metrics correlate weakly with user satisfaction scores, suggesting that user experience in voice-based AI systems depends on factors beyond technical performance. Our findings provide practical guidance for selecting components in multimodal conversations and contribute a validated evaluation methodology for human-AI interactions.

eess.AS↗

arXiv2025

Better Together: Quantifying the Benefits of AI-Assisted Recruitment

Hiring algorithms have mostly scored the materials recruiters already see. Large language models (LLMs) can instead generate new information about candidates by conducting, at scale, structured interviews once reserved for a few finalists. We study this shift in two field experiments at a recruitment platform. The first experiment holds the candidate pool fixed and randomizes whether recruiters observe the AI Interview Report; the second embeds the AI interview as a requirement in a live hiring pipeline. In both, candidates shortlisted with AI interview information pass the final human interview (conducted blind to shortlisting condition) at rates 17.5 (SE 8.5) to 20 (SE 11.8) percentage points higher than candidates shortlisted from resumes alone. The gains concentrate where resumes are least informative: adding AI Interview Report ratings to conventional candidate features raises out-of-sample AUC by 0.18 for junior candidates, against 0.08 for non-junior candidates. The participation cost falls on applicants as 75 percent of invited candidates do not complete the interview. However, the attrition is itself a signal: completion is more consistent with job-search motivation than with predicted interview performance. AI interviews thus add information exactly where conventional signals fail, and they move the cost of screening from firms to applicants.

cs.CL↗

arXiv2025

Zara: An LLM-based Candidate Interview Feedback System

This paper introduces Zara, an AI-driven recruitment support system developed by micro1, as a practical case study illustrating how large language models (LLMs) can enhance the candidate experience through personalized, scalable interview support. Traditionally, recruiters have struggled to deliver individualized candidate feedback due to logistical and legal constraints, resulting in widespread candidate dissatisfaction. Leveraging OpenAI's GPT-4o, Zara addresses these limitations by dynamically generating personalized practice interviews, conducting conversational AI-driven assessments, autonomously delivering structured and actionable feedback, and efficiently answering candidate inquiries using a Retrieval-Augmented Generation (RAG) system. To promote transparency, we have open-sourced the approach Zara uses to generate candidate feedback.

cs.HC↗

arXiv2025

RODEO: Robust Outlier Detection via Exposing Adaptive Out-of-Distribution Samples

In recent years, there have been significant improvements in various forms of image outlier detection. However, outlier detection performance under adversarial settings lags far behind that in standard settings. This is due to the lack of effective exposure to adversarial scenarios during training, especially on unseen outliers, leading to detection models failing to learn robust features. To bridge this gap, we introduce RODEO, a data-centric approach that generates effective outliers for robust outlier detection. More specifically, we show that incorporating outlier exposure (OE) and adversarial training can be an effective strategy for this purpose, as long as the exposed training outliers meet certain characteristics, including diversity, and both conceptual differentiability and analogy to the inlier samples. We leverage a text-to-image model to achieve this goal. We demonstrate both quantitatively and qualitatively that our adaptive OE method effectively generates ``diverse'' and ``near-distribution'' outliers, leveraging information from both text and image domains. Moreover, our experimental results show that utilizing our synthesized outliers significantly enhances the performance of the outlier detector, particularly in adversarial settings.

cs.CV↗

arXiv2025

Scanning Trojaned Models Using Out-of-Distribution Samples

Scanning for trojan (backdoor) in deep neural networks is crucial due to their significant real-world applications. There has been an increasing focus on developing effective general trojan scanning methods across various trojan attacks. Despite advancements, there remains a shortage of methods that perform effectively without preconceived assumptions about the backdoor attack method. Additionally, we have observed that current methods struggle to identify classifiers trojaned using adversarial training. Motivated by these challenges, our study introduces a novel scanning method named TRODO (TROjan scanning by Detection of adversarial shifts in Out-of-distribution samples). TRODO leverages the concept of "blind spots"--regions where trojaned classifiers erroneously identify out-of-distribution (OOD) samples as in-distribution (ID). We scan for these blind spots by adversarially shifting OOD samples towards in-distribution. The increased likelihood of perturbed OOD samples being classified as ID serves as a signature for trojan detection. TRODO is both trojan and label mapping agnostic, effective even against adversarially trained trojaned classifiers. It is applicable even in scenarios where training data is absent, demonstrating high accuracy and adaptability across various scenarios and datasets, highlighting its potential as a robust trojan scanning strategy.

cs.LG↗

arXiv2025

A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts

There have been several efforts to improve Novelty Detection (ND) performance. However, ND methods often suffer significant performance drops under minor distribution shifts caused by changes in the environment, known as style shifts. This challenge arises from the ND setup, where the absence of out-of-distribution (OOD) samples during training causes the detector to be biased toward the dominant style features in the in-distribution (ID) data. As a result, the model mistakenly learns to correlate style with core features, using this shortcut for detection. Robust ND is crucial for real-world applications like autonomous driving and medical imaging, where test samples may have different styles than the training data. Motivated by this, we propose a robust ND method that crafts an auxiliary OOD set with style features similar to the ID set but with different core features. Then, a task-based knowledge distillation strategy is utilized to distinguish core features from style features and help our model rely on core features for discriminating crafted OOD and ID sets. We verified the effectiveness of our method through extensive experimental evaluations on several datasets, including synthetic and real-world benchmarks, against nine different ND methods.

cs.CV↗

arXiv2021

MANA: Microarchitecting an Instruction Prefetcher

L1 instruction (L1-I) cache misses are a source of performance bottleneck. Sequential prefetchers are simple solutions to mitigate this problem; however, prior work has shown that these prefetchers leave considerable potentials uncovered. This observation has motivated many researchers to come up with more advanced instruction prefetchers. In 2011, Proactive Instruction Fetch (PIF) showed that a hardware prefetcher could effectively eliminate all of the instruction-cache misses. However, its enormous storage cost makes it an impractical solution. Consequently, reducing the storage cost was the main research focus in the instruction prefetching in the past decade. Several instruction prefetchers, including RDIP and Shotgun, were proposed to offer PIF-level performance with significantly lower storage overhead. However, our findings show that there is a considerable performance gap between these proposals and PIF. While these proposals use different mechanisms for instruction prefetching, the performance gap is largely not because of the mechanism, and instead, is due to not having sufficient storage. Prior proposals suffer from one or both of the following shortcomings: (1) a large number of metadata records to cover the potential, and (2) a high storage cost of each record. The first problem causes metadata miss, and the second problem prohibits the prefetcher from storing enough records within reasonably-sized storage.

cs.AR↗

arXiv2020

A Survey on Recent Hardware Data Prefetching Approaches with An Emphasis on Servers

Data prefetching, i.e., the act of predicting application's future memory accesses and fetching those that are not in the on-chip caches, is a well-known and widely-used approach to hide the long latency of memory accesses. The fruitfulness of data prefetching is evident to both industry and academy: nowadays, almost every high-performance processor incorporates a few data prefetchers for capturing various access patterns of applications; besides, there is a myriad of proposals for data prefetching in the research literature, where each proposal enhances the efficiency of prefetching in a specific way. In this survey, we discuss the fundamental concepts in data prefetching and study state-of-the-art hardware data prefetching approaches. Additional Key Words and Phrases: Data Prefetching, Scale-Out Workloads, Server Processors, and Spatio-Temporal Correlation.

cs.AR↗