Customer-obsessed science
Research areas
-
August 26, 20265 min readDiscounting the opinions of LLM judges with highly correlated outputs ensures that panels of judges reflect a true diversity of perspectives.
-
August 21, 20269 min read
-
July 30, 20268 min read
-
-
July 9, 202610 min read
Featured news
-
2026Continual learning methods for vision-language models are developed on benchmarks where each new task introduces entirely new domain knowledge. Real-world task sequences are more natural: they routinely share visual concepts, language patterns, and even training samples across stages. However, existing mixture-of-expert methods that assign one expert per task with fixed routing can split similar inputs
-
2026Time reasoning is a make-or-break capability for Large Language Models (LLMs) aspiring to act as reliable personal and enterprise assistants. This paper introduces the Temporal Reasoning Dataset (TRD), a programmatically generated multilingual benchmark designed to evaluate temporal reasoning operational capabilities in LLMs across ten languages, with particular focus on basic operations relevant to conversational
-
SIGMOD/PODS 20262026We present Aurora Limitless Database, a cloud-native distributed database system that extends Amazon Aurora PostgreSQL with horizontal scaling capabilities while maintaining strong consistency guarantees. The system provides transparent scalability using a router layer for query distribution and a storage layer of PostgreSQL shards, which eliminates the need for application-level sharding. Our key technical
-
ECML-PKDD 20262026Graph Neural Networks (GNNs) break down on zero-degree nodes, as message passing requires neighbors. Without interaction history, unseen entities are sub-optimally embedded, leaving them weakly anchored in the latent space, creating a cold-start bottleneck in retrieval. To address this, we propose GRAFT, a factorized architecture that unifies structural and feature transformations into a shared weight space
-
ECML-PKDD 20262026Accurate demand forecasting is vital for retail supply chain efficiency, yet a persistent trust-capacity gap limits industrial production to low-capacity interpretable models that fail to capture complex market dynamics. We propose Anchored FLoE, a dual-model framework that bridges this gap by fusing high-capacity deep learning with rigorous business guardrails. The framework integrates: (1) FLoE, an ensemble
Collaborations
View allWhether you're a faculty member or student, there are number of ways you can engage with Amazon.
View all