The JHU + Amazon Initiative for Interactive Artificial Intelligence (AI2AI) invites applications from outstanding PhD students for fellowships in the 2025-2026 academic year. Awardees, who will be designated Amazon Fellows, will receive a full stipend, 20% tuition, and student health insurance for the fall 2025 and spring 2026 semesters. Additionally, they will be nominated for a paid summer internship at Amazon in 2026, during which they will gain valuable industry insights and experiences via engagement with Amazon researchers.
Amazon builds and deploys AI across three technology layers: the bottom layer consists of Amazon’s own high performance and cost-effective custom chips, as well as a variety of other computing options including from third-parties. The middle layer focuses on the customer’s choice by providing the broadest selection of Foundation Models—both Amazon-built as well as those from other leading providers. At the top layer Amazon offers generative AI applications and services to improve every customer experience.
There are three things that distinguish Amazon’s approach to the development and deployment of AI:
1) Maintaining a strategic focus on improving the customer and employee experience through practical, real-world applications of AI;
2) marshaling world-class data, compute, and talent resources to drive AI innovation; and
3) committing to the development of responsible, reliable, and trustworthy AI.
Topics of interest include, but are not limited to, those below. Please feel free to bring your own unique viewpoint and expertise to these topics:
[dfd_spacer screen_wide_resolution=”1280″ screen_wide_spacer_size=”25″ screen_normal_resolution=”1024″ screen_tablet_resolution=”800″ screen_mobile_resolution=”480″ screen_normal_spacer_size=”25″ screen_tablet_spacer_size=”25″ screen_mobile_spacer_size=”25″][dfd_heading content_alignment=”text-left” delimiter_settings=”border-bottom-style:solid;|border-bottom-width:1px;|width:100px;|border-bottom-color:#fa9b25;” delimiter_margin=”margin-top:10px;margin-bottom:30px;” mobile_alignment=”left” style=”style_01″ heading_margin=”margin-bottom:10px;” title_font_options=”tag:h4|color:%23ffffff” subtitle_font_options=”tag:h3″]Topics[/dfd_heading][dfd_icon_list del_height=”1″ del_style=”solid” del_color=”#000000″ main_style=”style-3″ font_options=”font_size:16|color:%23ffffff”][dfd_icon_list_item icon_type=”none”]
Foundational Model Improvements:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Novel model architectures[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Novel training algorithms and methodologies[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Multi-modal (e.g., text, image, video, audio) understanding and generation[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Multi-lingual understanding and generation[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Algorithms and workflows for acquiring and curating high-quality and diverse datasets for training[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Foundation Model Evaluation:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Creation of new benchmarks for assessing foundation model capabilities[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Methodologies for robust evaluation of generative AI systems, including agents[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Efficient Generative AI:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Efficient training and inference for cloud and on-device applications[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Compute and memory efficient handling of multi-modal long/infinite context[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Improving efficiency of diffusion models, including discrete diffusion[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Reasoning:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Commonsense and domain-specific (e.g., math, coding) reasoning[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Temporal and spatial reasoning[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Reasoning for planning[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Test-time scaling[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Creative problem solving (learning out-of-the-box thinking techniques)[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Knowledge Grounding:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Approaches to ground (multi-modal) generation on up-to-date world, domain-specific, enterprise, or personal knowledge[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Memory-augmented generative AI systems[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]AI truthfulness and hallucination reduction[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Agentic AI:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Creation of autonomous systems capable of performing tasks, making decisions, and interacting with their environments[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Multi-Agent systems and agent orchestration framework improvements[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Customization and continual improvement of agents post deployment[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Internationalizing agentic systems[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Responsible Generative AI:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Red teaming (e.g., advanced red teaming approaches, automated red teaming for multi-modal models)[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Improvement of foundation model RAI performance (e.g., robustness to jailbreaking and membership influence attacks; watermarking approaches; deepfake detection)[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Responsible agentic AI (e.g., robustness of multi-agent systems, adherence to guardrails)[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Privacy-preserving Generative AI[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Interpretable AI:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Understanding mechanisms behind emergent AI capabilities (e.g. mechanistic interpretation)[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Communicating AI agents’ rationales and search process (meta-cognition) with human stakeholders[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Small-size LLM:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]LLM distillation[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]LLM customization for specific skills[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Knowledge-graph aware LLM[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Long-context, small-size LLM customized for reasoning capability[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Applications of Generative AI:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Systems that leverage Generative AI for advancing science and technology in areas such as physics, mathematics, chemistry, biology, hardware design, materials science, engineering, economics, healthcare, climate[/dfd_icon_list_item][dfd_icon_list_item icon_type=”none”]
Security:[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Low-cost computer vision techniques for object interactions for physical security[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Near-real-time anomaly detection at scale (e.g., streaming petabytes per hour) to identify malicious activity in Linux cloud-based environments[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Running distributed big data, AI/ML, and/or streaming models with minimal latency and cost using host, network, and audit telemetry sources. Such data engineering can be used for selecting interesting subsets of data including complex, multi-event sequences, improving matching and micro-summarization efficiencies for continuous log processing, and improving event summarization processing efficiency while maintaining accuracy[/dfd_icon_list_item][dfd_icon_list_item icon=”dfd-icon-right_6 dfd_icon_set-icon-right_6″]Use of AI/ML (e.g., GenAI) to automate and improve security response workflows (for example, from written reports or threat detections), specifically through incident summarization, response planning, and communication management[/dfd_icon_list_item][/dfd_icon_list][dfd_spacer screen_wide_resolution=”1280″ screen_wide_spacer_size=”25″ screen_normal_resolution=”1024″ screen_tablet_resolution=”800″ screen_mobile_resolution=”480″ screen_normal_spacer_size=”25″ screen_tablet_spacer_size=”25″ screen_mobile_spacer_size=”25″]
[dfd_spacer screen_wide_resolution=”1280″ screen_wide_spacer_size=”40″ screen_normal_resolution=”1024″ screen_tablet_resolution=”800″ screen_mobile_resolution=”480″ screen_normal_spacer_size=”25″ screen_tablet_spacer_size=”25″ screen_mobile_spacer_size=”25″][dfd_heading content_alignment=”text-left” delimiter_settings=”border-bottom-style:solid;|border-bottom-width:1px;|width:100px;|border-bottom-color:#fa9b25;” delimiter_margin=”margin-top:10px;margin-bottom:30px;” mobile_alignment=”left” style=”style_01″ heading_margin=”margin-bottom:10px;” title_font_options=”tag:h4|color:%23ffffff” subtitle_font_options=”tag:h3″]Eligibility[/dfd_heading]
Applicants must be enrolled full time and in good standing in a WSE PhD program or PhD program in a WSE affiliated department, be in their third year or higher in AY 2025-2026, and have exhibited outstanding academic performance. Exceptionally outstanding students in their second year in AY 2025-2026 also are eligible. Applications from those working in AI who are women and/or identify as members of underrepresented groups are particularly encouraged.
[dfd_spacer screen_wide_resolution=”1280″ screen_wide_spacer_size=”25″ screen_normal_resolution=”1024″ screen_tablet_resolution=”800″ screen_mobile_resolution=”480″ screen_normal_spacer_size=”25″ screen_tablet_spacer_size=”25″ screen_mobile_spacer_size=”25″]
