arXiv subjects

Rocky Slavin

Publications and source records attributed to Rocky Slavin.

8 recordsLinked to original sources

arXiv2026

TRACER: Early Failure Detection for Task-Oriented Dialogue

Task-oriented dialogue systems often fail before the final breakdown is obvious, but most evaluation only measures failure after the conversation has already gone wrong. We present TRACER, a method for early failure detection in task-oriented dialogue. TRACER predicts from a partial dialogue whether the full conversation will eventually fail by combining simple trajectory signals from belief-state changes with text representations of the evolving dialogue state. We evaluate the method in both oracle and generated belief-state settings, and test how well it works when only 25%, 50%, 75%, or 100% of the dialogue is visible. Across these settings, TRACER detects useful failure signals well before the end of the conversation and outperforms heuristic, classical, and single-stream baselines. These results suggest that early failure detection can provide a practical warning signal for dialogue systems before the interaction fully breaks down.

cs.CL↗

arXiv2026

When Retrieval Doesn't Help: A Large-Scale Study of Biomedical RAG

Medical question answering is a high-stakes setting where factual errors can have serious consequences. Retrieval-augmented generation (RAG) is widely viewed as a promising solution, and prior work has reported substantial gains for large medical QA models. We revisit this assumption across a broad range of open-weight instruction-tuned models spanning 7B to 72B parameters. Across five models, ten biomedical QA datasets, four retrieval methods, and four retrieval corpora, we find that retrieval yields only small and inconsistent improvements over a no-retrieval baseline, typically within 1-2 points. In contrast, the choice of backbone model has a much larger effect than the choice of retriever or corpus, and expert and layman retrieval sources perform similarly in most settings. These results suggest that the main bottleneck is not retrieval quality alone, but the model's limited ability to use retrieved evidence effectively.

cs.CL↗

arXiv2026

CogAdapt: Adapting Clinical ECG Foundation Models for Wearable Cognitive Load Assessment

Assessing cognitive load continuously and at low latency would help adaptive human-computer interaction, but it remains hard because labeled data are scarce and models generalize poorly across subjects. Recent ECG foundation models, pre-trained on millions of clinical diagnostic ECG recordings, yet they do not apply directly to wearable devices when the sensor configuration and the task both differ. We present CogAdapt, a framework that adapts a clinical ECG foundation model to wearable cognitive load assessment. CogAdapt has two parts. LeadBridge is a learnable adapter that maps 3-lead wearable signals to a 12-lead-compatible representation. ProFine is a progressive fine-tuning strategy that unfreezes encoder layers in stages while limiting representational drift in the pre-trained model. On two public datasets (CLARE and CL-Drive) under leave-one-subject-out cross-validation, CogAdapt reaches macro-F1 of 0.626 and 0.768, improving over from-scratch baselines by 11.2 and 16.1 percentage points. The results show that a clinical ECG pretraining can support subject-independent cognitive load assessment from wearable sensors.

cs.LG↗

arXiv2026

MambaGaze: Bidirectional Mamba with Explicit Missing Data Modeling for Cognitive Load Assessment from Eye-Gaze Tracking Data

Real-time cognitive load assessment from eye-tracking signals could enable adaptive human-centered AI in safety-critical applications such as driver vigilance monitoring or automated flight deck assistance, yet two challenges persist: handling frequent data missingness from blinks and tracking failures, and efficiently modeling long-range temporal dependencies. We propose MambaGaze (Bi-Mamba), a framework that addresses these challenges through (1) XMD encoding, which augments raw features with observation masks and time-deltas to explicitly model data uncertainty, and (2) bidirectional Mamba-2, which captures temporal dependencies with linear computational complexity. Experiments on CLARE and CL-Drive datasets under leave-one-subject-out evaluation show that MambaGaze achieves 77.1% accuracy and 59.2% macro-F1 on CLARE, and 69.4% accuracy and 51.5% macro-F1 on CL-Drive, attaining the highest average LOSO macro-F1 (55.3%) across all ten compared models. Input-stream ablation indicates that log-scaled time-deltas are the strongest single channel in our setting, and combining all three XMD streams provides consistent gains of 5-20 pp macro-F1. Edge deployment benchmarks on three NVIDIA Jetson Orin platforms show real-time inference at 27-36 FPS with power consumption below 6.6 W, supporting feasibility for embedded cognitive load monitoring.

cs.LG↗

arXiv2026

Prompting Underestimates LLM Capability for Time Series Classification

Prompt-based evaluations suggest that large language models (LLMs) perform poorly on time series classification, raising doubts about whether they encode meaningful temporal structure. We show that this conclusion reflects limitations of prompt-based generation rather than the model's representational capacity by directly comparing prompt outputs with linear probes over the same internal representations. While zero-shot prompting performs near chance, linear probes improve average F1 from 0.15-0.26 to 0.61-0.67, often matching or exceeding specialized time series models. Layer-wise analyses further show that class-discriminative time series information emerges in early transformer layers and is amplified by visual and multimodal inputs. Together, these results demonstrate a systematic mismatch between what LLMs internally represent and what prompt-based evaluation reveals, leading current evaluations to underestimate their time series understanding.

cs.CL↗

arXiv2024

An Analysis of Automated Use Case Component Extraction from Scenarios using ChatGPT

Mobile applications (apps) are often developed by only a small number of developers with limited resources, especially in the early years of the app's development. In this setting, many requirements acquisition activities, such as interviews, are challenging or lower priority than development and release activities. Moreover, in this early period, requirements are frequently changing as mobile apps evolve to compete in the marketplace. As app development companies move to standardize their development processes, however, they will shift to documenting and analyzing requirements. One low-cost source of requirements post-deployment are user-authored scenarios describing how they interact with an app. We propose a method for extracting use case components from user-authored scenarios using large language models (LLMs). The method consists of a series of prompts that were developed to improve precision and recall on a ground truth dataset of 50 scenarios independently labeled with UC components. Our results reveal that LLMs require additional domain knowledge to extract UC components, and that refining prompts to include this knowledge improves the quality of the extracted UC components.

cs.SE↗

arXiv2024

Toward Regulatory Compliance: A few-shot Learning Approach to Extract Processing Activities

The widespread use of mobile applications has driven the growth of the industry, with companies relying heavily on user data for services like targeted advertising and personalized offerings. In this context, privacy regulations such as the General Data Protection Regulation (GDPR) play a crucial role. One of the GDPR requirements is the maintenance of a Record of Processing Activities (RoPA) by companies. RoPA encompasses various details, including the description of data processing activities, their purposes, types of data involved, and other relevant external entities. Small app-developing companies face challenges in meeting such compliance requirements due to resource limitations and tight timelines. To aid these developers and prevent fines, we propose a method to generate segments of RoPA from user-authored usage scenarios using large language models (LLMs). Our method employs few-shot learning with GPT-3.5 Turbo to summarize usage scenarios and generate RoPA segments. We evaluate different factors that can affect few-shot learning performance consistency for our summarization task, including the number of examples in few-shot learning prompts, repetition, and order permutation of examples in the prompts. Our findings highlight the significant influence of the number of examples in prompts on summarization F1 scores, while demonstrating negligible variability in F1 scores across multiple prompt repetitions. Our prompts achieve successful summarization of processing activities with an average 70% ROUGE-L F1 score. Finally, we discuss avenues for improving results through manual evaluation of the generated summaries.

cs.SE↗

arXiv2024

Beyond Text-to-SQL for IoT Defense: A Comprehensive Framework for Querying and Classifying IoT Threats

Recognizing the promise of natural language interfaces to databases, prior studies have emphasized the development of text-to-SQL systems. While substantial progress has been made in this field, existing research has concentrated on generating SQL statements from text queries. The broader challenge, however, lies in inferring new information about the returned data. Our research makes two major contributions to address this gap. First, we introduce a novel Internet-of-Things (IoT) text-to-SQL dataset comprising 10,985 text-SQL pairs and 239,398 rows of network traffic activity. The dataset contains additional query types limited in prior text-to-SQL datasets, notably temporal-related queries. Our dataset is sourced from a smart building's IoT ecosystem exploring sensor read and network traffic data. Second, our dataset allows two-stage processing, where the returned data (network traffic) from a generated SQL can be categorized as malicious or not. Our results show that joint training to query and infer information about the data can improve overall text-to-SQL performance, nearly matching substantially larger models. We also show that current large language models (e.g., GPT3.5) struggle to infer new information about returned data, thus our dataset provides a novel test bed for integrating complex domain-specific reasoning into LLMs.

cs.CL↗