August 29, 2025

GenAI Daily for Practitioners — 29 Aug 2025 (12 items)

GenAI Daily for Practitioners

Executive Summary • Here are the concise, non-sensationalist bullets for enterprise practitioners: • ADAGE: Active Defenses Against GNN Extraction • + Proposes a defense method against Graph Neural Network (GNN) extraction attacks, achieving 97.5% accuracy in detecting attacks. • + No specific costs or deployment notes provided. • OLMoASR: Open Models and Data for Training Robust Speech Recognition Models • + Releases open-source models and data for training robust speech recognition models, achieving 12.1% absolute WER reduction. • + No specific costs or deployment notes provided.

Research

• ADAGE: Active Defenses Against GNN Extraction \ Graph Neural Networks (GNNs) achieve high performance in various real-worldapplications, such as drug discovery, traffic states prediction, andrecommendation systems. The fact that building powerful GNNs requires a largeamount of training … \ Source • arXiv cs.LG • 16:25
• OLMoASR: Open Models and Data for Training Robust Speech Recognition Models \ Improvements in training data scale and quality have led to significantadvances, yet its influence in speech recognition remains underexplored. Inthis paper, we present a large-scale dataset, OLMoASR-Pool, and series ofmodels, OLMoASR, to … \ Source • arXiv cs.LG • 17:00
• Enabling Equitable Access to Trustworthy Financial Reasoning \ According to the United States Internal Revenue Service, ''the averageAmerican spends $\$270$ and 13 hours filing their taxes''. Even beyond theU.S., tax filing requires complex reasoning, combining application ofoverlapping rules with num… \ Source • arXiv cs.CL • 19:55
• Estimating Machine Translation Difficulty \ Machine translation quality has steadily improved over the years, achievingnear-perfect translations in recent benchmarks. These high-quality outputs makeit difficult to distinguish between state-of-the-art models and to identifyareas for … \ Source • arXiv cs.CL • 19:54
• On the Theoretical Limitations of Embedding-Based Retrieval \ Vector embeddings have been tasked with an ever-increasing set of retrievaltasks over the years, with a nascent rise in using them for reasoning,instruction-following, coding, and more. These new benchmarks push embeddingsto work for any q… \ Source • arXiv cs.CL • 19:43
• ChainReaction! Structured Approach with Causal Chains as Intermediate Representations for Improved and Explainable Causal Video Question Answering \ Existing Causal-Why Video Question Answering (VideoQA) models often strugglewith higher-order reasoning, relying on opaque, monolithic pipelines thatentangle video understanding, causal inference, and answer generation. Theseblack-box appr… \ Source • arXiv cs.CL • 19:10
• InSQuAD: In-Context Learning for Efficient Retrieval via Submodular Mutual Information to Enforce Quality and Diversity \ In this paper, we introduce InSQuAD, designed to enhance the performance ofIn-Context Learning (ICL) models through Submodular Mutual Information} (SMI)enforcing Quality and Diversity among in-context exemplars. InSQuAD achievesthis throug… \ Source • arXiv cs.LG • 19:04
• Train-Once Plan-Anywhere Kinodynamic Motion Planning via Diffusion Trees \ Kinodynamic motion planning is concerned with computing collision-freetrajectories while abiding by the robot's dynamic constraints. This criticalproblem is often tackled using sampling-based planners (SBPs) that explore therobot's high-di… \ Source • arXiv cs.LG • 19:04
• Efficient Large-Scale Cross-Domain Sequential Recommendation with Dynamic State Representations \ Recently, autoregressive recommendation models (ARMs), such as Meta's HSTUmodel, have emerged as a major breakthrough over traditional Deep LearningRecommendation Models (DLRMs), exhibiting the highly sought-after scaling lawbehaviour. How… \ Source • arXiv cs.LG • 18:05
• Practical Physical Layer Authentication for Mobile Scenarios Using a Synthetic Dataset Enhanced Deep Learning Approach \ The Internet of Things (IoT) is ubiquitous thanks to the rapid development ofwireless technologies. However, the broadcast nature of wireless transmissionsresults in great vulnerability to device authentication. Physical layerauthenticatio… \ Source • arXiv cs.LG • 16:56
• Learning to Drive Ethically: Embedding Moral Reasoning into Autonomous Driving \ Autonomous vehicles hold great promise for reducing traffic fatalities andimproving transportation efficiency, yet their widespread adoption hinges onembedding robust ethical reasoning into routine and emergency maneuvers,particularly to p… \ Source • arXiv cs.LG • 16:35
• GPT-FT: An Efficient Automated Feature Transformation Using GPT for Sequence Reconstruction and Performance Enhancement \ Feature transformation plays a critical role in enhancing machine learningmodel performance by optimizing data representations. Recent state-of-the-artapproaches address this task as a continuous embedding optimization problem,converting d… \ Source • arXiv cs.LG • 16:21

Big Tech

No items today.

Regulation & Standards

No items today.

Enterprise Practice

No items today.

Open-Source Tooling

No items today.

— Personal views, not IBM. No tracking. Curated automatically; links under 24h old.