Suchi Saria

Open-Set Domain Adaptation Under Background Distribution Shift: Challenges and A Provably Efficient Solution

Closed-Form Reliability Modeling for Distributed AI Inference on Volatile Consumer Edge Devices

This paper introduces an analytical framework for quantifying computational reliability in Extreme Edge Computing (XEC), where AI inference workloads are distributed across consumer-owned devices with unpredictable availability. The framework derives closed-form reliability expressions — the probability that instantaneous device capacity meets a QoS demand threshold — under both minimal-information and historical-data regimes. It extends to multi-device topologies (series, parallel, partitioned) and provides optimal workload allocation rules, giving orchestrators tractable tools for deployment feasibility assessment. Validation against YOLO11m-based object detection on emulated XEC environments shows strong agreement between analytical predictions, Monte Carlo simulation, and empirical measurements.

Massively Multimodal Foundation Models: A Framework for Capturing Interactions with Specialized Mixture-of-Experts

Open-Set Domain Adaptation Under Background Distribution Shift: Challenges and A Provably Efficient Solution

On the Invariance and Generality of Neural Scaling Laws

FLAME: Adaptive Mixture-of-Experts for Continual Multimodal Multi-Task Learning