Grant Opportunities

The Amazon Research Awards fall cycle is now open, and the deadline is November 4th. Awards include cash up to $50,000-$80,000 and AWS credits up to $40,000-$50,000. 

Here are the categories of interest for this cycle:

Supports research and open-source tools for AI agents, including low/no-code agent development, human-AI collaboration, multi-agent systems, agent customization, verification, safety/governance, and agentic AI for accelerating scientific discovery, including physics.

Topics include trustworthy security agents, threat/anomaly detection, agent access governance, AI-driven incident response and red-teaming, vulnerability detection, secure foundation models, reinforcement learning for security, graph anomaly detection, learning from noisy/limited labels, malware analysis, and secure code generation.

More operationally focused than the preceding CFP. Topics include self-improving vulnerability detection, security agents that recognize when human intervention is necessary, computer vision for physical security, very-large-scale anomaly detection, efficient analysis of streaming telemetry, automated red-team/blue-team systems, and monitoring/containment of autonomous agents.

Supports formal methods and mathematical reasoning for trustworthy computing, including neurosymbolic reasoning combining LLMs with solvers/proof checking, theorem-prover interfaces, automated testing, formal hardware models, hardware/software verification, and formal verification of neural-network robustness—particularly object detection.

Covers fundamental and applied cryptography, including the feasibility of cryptographically relevant quantum computers, fully homomorphic encryption (FHE), multiparty computation, cryptographic software/tools, privacy-preserving distributed collaboration, and AI security. AI topics include provenance/watermarking, differential privacy, privacy-preserving AI training/inference, and formal definitions of AI-security concepts.

Focuses on AI agents that can generate, optimize, test and iteratively improve the software underlying AI workloads on AWS Trainium/Inferentia hardware. Topics include kernel generation/optimization, compile-execute-repair loops, auto-tuning, reinforcement learning from compiler/profiler feedback, long-horizon agent learning and self-improving AI systems.

Supports datasets, benchmarks and evaluation methods for underrepresented and low-resource languages, culturally contextual AI evaluation, responsible/safe multilingual AI, cross-lingual transfer, and research on AI deployment in emerging markets, with particular attention to the Global South.

Targets perception on constrained devices with limited optics, power and computing. Three principal topics are generative visual reconstruction from incomplete observations; efficient visual encoding for vision-language models; and learned multimodal sensor fusion for object tracking. The last includes visual, inertial and ranging data, self-supervised fusion, calibration-free deployment and tracking through sensor occlusion.

Seeks alternatives to costly/privacy-sensitive real-world multimodal training data. Four areas are text-to-simulation human motion/object interaction; causally coherent audio-visual generation; long-horizon multimodal episode generation; and action-conditioned scenario synthesis/world models.

Please share with others who might be interested! If you or your colleagues have any questions or would like assistance with your proposal, please reach out to our office. If you would like proposal development or budget/routing assistance, please fill out the Research Development Services Request Form.