Amazon Science

Expo Talk: Strands robots: Unifying robot control, simulation, and training behind natural language July 7, 12:00 AM - 1:00 AM KST Room: Hall B2 Speakers: Cagatay Cali, Yin Song Abstract: Despite rapid advances in vision-language-action (VLA) models, deploying robot intelligence remains fragmented: different SDKs for different robots, different policy frameworks with incompatible interfaces, an unbridged simulation-to-reality gap, and training pipelines that demand specialist expertise .

At Amazon, we believe that as our strategy to reach net-zero by 2040 evolves, we need to continue to raise the bar on what and how we measure. Measuring carbon emissions across our entire business is complex, and the tools and methodologies available are continuously improving.

As an Applied Scientist, you will be responsible for bringing new product designs through to manufacturing. You will work closely with multi-disciplinary groups including Product Design, Industrial Design, Hardware Engineering, and Operations, to drive key aspects of engineering of consumer electronics products.

E-commerce assistants must go beyond product search to support idea inspiration, criteria formation, comparison, and tool-grounded fact-checking over non-linear shopping journeys. Teaching these behaviors into deployable latency-constrained models is bottlenecked by post-training data: trajectories must cover the full agentic workflow with diversity and fidelity, yet desired outputs are open-ended (often 'essay-like') without a single ground truth to ensure quality.

What happens when you combine startup speed with Amazon-scale impact? You get this team. Amazon Enterprise Security Products is a newly launched group building intelligent, cloud-agnostic security tools using AI-first development practices. Here, you build AI and you build with AI at the same time. This role is a chance to define and lead the science strategy for the future of security tooling with a small, fast team that ships like a startup but deploys at Amazon scale.

Tao Tu, Qing Ping, Govind Thattai, Gokhan Tur, Prem Natarajan Copy BibTeX GuessWhat?! is a visual dialog guessing game which incorporates a Questioner agent that generates a sequence of questions, while an Oracle agent answers the respective questions about a target object in an image. Based on this dialog history between the Questioner and the Oracle, a Guesser agent makes a final guess of the target object.

Large-scale AI evaluation increasingly relies on aggregating binary judgments from K annotators, including LLMs used as judges. Most classical methods, e.g., Dawid-Skene or (weighted) majority voting, assume annotators are conditionally independent given the true label Y ∈ {0, 1}, an assumption often violated by LLM judges due to shared data, architectures, prompts, and failure modes. Ignoring such dependencies can yield miscalibrated posteriors and even confidently incorrect predictions.

Yuefeng Peng, Parnian Afshar, Megan Ganji, Thomas Butler, Amir Houmansadr, Mingxian Wang, Dezhi Hong Copy BibTeX Large language models can memorize information that must be removed–ranging from copyright-sensitive content (e.g., book chapters) to personally identifiable information (e.g., income)–to ensure responsible and compliant behavior. Unlearning has emerged as an efficient alternative to full retraining, aiming to remove specific knowledge.

Automatic evaluation of retrieval augmented generation (RAG) systems relies on fine grained dimensions like faithfulness and relevance, as judged by expert human annotators. Meta-evaluation benchmarks support the development of automatic evaluators that correlate well with human judgement. However, existing benchmarks predominantly focus on English or use translated data, which fails to capture cultural nuances. A native approach provides a better representation of the end user experience.

Recent agentic approaches to LLM-based kernel generation have achieved strong results on CUDA, yet emerging AI accelerators such as AWS Trainium and Inferentia remain unaddressed. Writing kernels for these chips via the Neuron Kernel Interface (NKI) is particularly challenging due to a multi-engine architecture, tile-based programming with a fixed 128-element partition dimension, and explicit memory management. Moreover, no training data, benchmarks, or tool-augmented agents exist for this domain.