Vijay Janapa Reddi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Vijay Janapa Reddi has a total of 245 co-authors. Here are the top co-authors by number of shared publications:
- Jason Jabbour
Harvard University, Boston, Massachusetts, United States
5 shared publications - Zishen Wan
Georgia Institute of Technology, Atlanta, GA, USA
5 shared publications - Alicia Parrish
New York University
3 shared publications - Arijit Raychowdhury
Georgia Institute of Technology
3 shared publications - Arya Tschand
Harvard University
3 shared publications - Brian Plancher
Barnard College, Columbia University, New York, NY, USA
3 shared publications
Latest Publications
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 79-86
MLPerf Power: Addressing the Artificial Intelligence Power Wall
Power has become the defining constraint for modern machine learning, from milliwatt edge devices to multi-megawatt training clusters. Across this nine-order-of-magnitude span, the field has lacked a consistent way to measure and compare energy efficiency. We present MLPerf Power, a cross-scale energy benchmarking methodology that adapts instrumentation while preserving common measurement principles and reports shared metrics such as samples per joule and tokens per joule. The methodology has produced 1,841 reproducible submissions from 60 systems. The corpus shows that easy efficiency gains are plateauing, that energy scales nonlinearly with accelerator count, and that low-precision innovation has substantially reduced the cost of high-accuracy inference. Extending to LLM inference, we show that sparse mixture-of-experts models pay a memory-bandwidth penalty that contradicts training intuitions. The publicly accessible corpus already informs industry roadmaps, supports infrastructure planning, and provides ground-truth telemetry for energy-aware optimization and verified carbon accounting in AI systems.
- 1 September 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (5) , 79-86
MLPerf Power: Addressing the Artificial Intelligence Power Wall
Power has become the defining constraint for modern machine learning, from milliwatt edge devices to multi-megawatt training clusters. Across this nine-order-of-magnitude span, the field has lacked a consistent way to measure and compare energy efficiency. We present MLPerf Power, a cross-scale energy benchmarking methodology that adapts instrumentation while preserving common measurement principles and reports shared metrics such as samples per joule and tokens per joule. The methodology has produced 1,841 reproducible submissions from 60 systems. The corpus shows that easy efficiency gains are plateauing, that energy scales nonlinearly with accelerator count, and that low-precision innovation has substantially reduced the cost of high-accuracy inference. Extending to LLM inference, we show that sparse mixture-of-experts models pay a memory-bandwidth penalty that contradicts training intuitions. The publicly accessible corpus already informs industry roadmaps, supports infrastructure planning, and provides ground-truth telemetry for energy-aware optimization and verified carbon accounting in AI systems.
- Open Access
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 2608-2642
Going PLACES: Participatory Localized Red Teaming for Text-to-Image Safety in the Global South
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 2608-2642
Open AccessGoing PLACES: Participatory Localized Red Teaming for Text-to-Image Safety in the Global South
- Open Access
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 7422-7449
Bridging the Scale Gap: Augmenting Human Red-Teaming to Uncover Latent Risks in T2I Models
- 25 June 2026
- Published by Association for Computing Machinery (ACM)
- p. 7422-7449
Open AccessBridging the Scale Gap: Augmenting Human Red-Teaming to Uncover Latent Risks in T2I Models
- Open Access
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1089-1112
Lifetime-Aware Design for Item-Level Intelligence at the Extreme Edge
- 22 March 2026
- Published by Association for Computing Machinery (ACM)
- p. 1089-1112
Open AccessLifetime-Aware Design for Item-Level Intelligence at the Extreme Edge
- 23 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (3) , 77-87
Breaking Tiny Machine Learning: Why Quantized Neural Networks Need Domain-Specific Security Analysis
Most TinyML hardware focus on supporting Quantized Neural Networks (QNNs) to meet stringent constraints on power consumption, size, and cost. Despite this, the security aspects of quantization within TinyML hardware remain largely unexplored. Although previous studies indicate that QNNs demonstrate similar or enhanced robustness when compared to full-precision Deep Neural Networks (DNNs) against typical evasion attacks, no attack strategies tailored specifically for TinyML hardware have been proposed yet. This paper addresses the aforementioned shortfall by demonstrating how a two-step attack pipeline can surpass the current state-of-the-art in the QNN context and shows the need for more hardware-aware security research.
- 23 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Micro
- Vol. 46 (3) , 77-87
Breaking Tiny Machine Learning: Why Quantized Neural Networks Need Domain-Specific Security Analysis
Most TinyML hardware focus on supporting Quantized Neural Networks (QNNs) to meet stringent constraints on power consumption, size, and cost. Despite this, the security aspects of quantization within TinyML hardware remain largely unexplored. Although previous studies indicate that QNNs demonstrate similar or enhanced robustness when compared to full-precision Deep Neural Networks (DNNs) against typical evasion attacks, no attack strategies tailored specifically for TinyML hardware have been proposed yet. This paper addresses the aforementioned shortfall by demonstrating how a two-step attack pipeline can surpass the current state-of-the-art in the QNN context and shows the need for more hardware-aware security research.
- Top Cited PapersOpen Access
- 21 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
TinyML Security: Attacks, Defenses, and Open Challenges in Resource-Constrained Machine Learning Systems
Tiny Machine Learning (TinyML) enables machine learning inference on microcontrollers with kilobytes of memory and megahertz processors, two to three orders of magnitude more constrained than conventional edge devices. These extreme limitations render traditional security measures impractical, yet the security implications remain underexplored. Our systematic literature review reveals that fewer than 5% of TinyML publications address security concerns, with even fewer focusing on TinyML-specific vulnerabilities, a critical gap as these devices proliferate in safety-critical applications including healthcare monitoring, autonomous systems, and industrial control. This paper provides a comprehensive security survey specifically addressing TinyML’s resource constraints and unique threat landscape. We develop a resource-based device taxonomy distinguishing IoT, EdgeML, and TinyML security capabilities; formulate a TinyML-specific threat model identifying physical and remote attack vectors; systematically analyze eleven attack classes across hardware, software, and model layers; and assess threat severity using the Common Vulnerability Scoring System (CVSS). For each attack, we evaluate whether conventional countermeasures are feasible under TinyML constraints by assessing computational overhead, memory requirements, and practical deployability on representative platforms. Our analysis reveals critical gaps where existing defenses impose prohibitive overhead, requiring new lightweight solutions. We conclude by identifying open research challenges specific to securing resource-constrained machine learning systems, providing a roadmap for future work.
- 21 February 2026
- Published by Association for Computing Machinery (ACM) in ACM Computing Surveys
Top Cited PapersOpen AccessTinyML Security: Attacks, Defenses, and Open Challenges in Resource-Constrained Machine Learning Systems
Tiny Machine Learning (TinyML) enables machine learning inference on microcontrollers with kilobytes of memory and megahertz processors, two to three orders of magnitude more constrained than conventional edge devices. These extreme limitations render traditional security measures impractical, yet the security implications remain underexplored. Our systematic literature review reveals that fewer than 5% of TinyML publications address security concerns, with even fewer focusing on TinyML-specific vulnerabilities, a critical gap as these devices proliferate in safety-critical applications including healthcare monitoring, autonomous systems, and industrial control. This paper provides a comprehensive security survey specifically addressing TinyML’s resource constraints and unique threat landscape. We develop a resource-based device taxonomy distinguishing IoT, EdgeML, and TinyML security capabilities; formulate a TinyML-specific threat model identifying physical and remote attack vectors; systematically analyze eleven attack classes across hardware, software, and model layers; and assess threat severity using the Common Vulnerability Scoring System (CVSS). For each attack, we evaluate whether conventional countermeasures are feasible under TinyML constraints by assessing computational overhead, memory requirements, and practical deployability on representative platforms. Our analysis reveals critical gaps where existing defenses impose prohibitive overhead, requiring new lightweight solutions. We conclude by identifying open research challenges specific to securing resource-constrained machine learning systems, providing a roadmap for future work.
- 1 January 2026
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 3 (1)
International Retrospective Observational Study of Continual Learning for AI on Endotracheal Tube Placement from Chest Radiographs
- 1 January 2026
- Published by Massachusetts Medical Society in NEJM AI
- Vol. 3 (1)
International Retrospective Observational Study of Continual Learning for AI on Endotracheal Tube Placement from Chest Radiographs
- Open Access
- 21 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
SocratiQ: A Generative AI-Powered Learning Companion for Personalized Education and Broader Accessibility
- 21 June 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-10
Open AccessSocratiQ: A Generative AI-Powered Learning Companion for Personalized Education and Broader Accessibility
- Open Access
- 8 May 2025
- Published by Association for Computing Machinery (ACM) in Queue
- Vol. 23 (2) , 79-137
Generative AI at the Edge: Challenges and Opportunities
Generative AI at the edge is the next phase in AI's deployment: from centralized supercomputers to ubiquitous assistants and creators operating alongside humans. The challenges are significant but so are the opportunities for personalization, privacy, and innovation. By tackling the technical hurdles and establishing new frameworks (conceptual and infrastructural), we can ensure this transition is successful and beneficial.
- 8 May 2025
- Published by Association for Computing Machinery (ACM) in Queue
- Vol. 23 (2) , 79-137
Open AccessGenerative AI at the Edge: Challenges and Opportunities
Generative AI at the edge is the next phase in AI's deployment: from centralized supercomputers to ubiquitous assistants and creators operating alongside humans. The challenges are significant but so are the opportunities for personalization, privacy, and innovation. By tackling the technical hurdles and establishing new frameworks (conceptual and infrastructural), we can ensure this transition is successful and beneficial.
- Open Access
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 982-997
ReCA: Integrated Acceleration for Real-Time and Efficient Cooperative Embodied Autonomous Agents
- 30 March 2025
- Published by Association for Computing Machinery (ACM)
- p. 982-997
Open AccessReCA: Integrated Acceleration for Real-Time and Efficient Cooperative Embodied Autonomous Agents