Reinforcement Learning for Control
Multi-agent and multi-objective high-assurance RL for physical systems.
Machine Learning · Reinforcement Learning · LLM Agents . Digital Twins
Soumyendu Sarkar
Senior Director & Senior Distinguished Technologist, Leading ML Research @ Hewlett Packard Enterprise Lab.
Research Interests
AI methods grounded in physical and operational constraintsDigital Twins for Infrastructure
Liquid-cooled data centers, HPC, and wave energy.
LLM Agents for Operations
Agentic AI copilots & deep reasoning, and Tokenomics.
Sustainable AI
Resource-aware ML / Data Center holistic optimization.
AI for Science
Fusion, optimization, inverse design, generative ML & AI for Quantum.
Trustworthy & High-Assurance AI
Autonomy, and explainable AI.
Recent Top AI Conference Papers
Top-conference papers as first or principal author
Recent work emphasizes trustworthy AI, reinforcement learning, sustainable data centers, digital twins, liquid cooling, and scientific ML.
NeurIPS 2026Data Center - Energy GridC2G-BENCH: A Cyber-Physical Evaluation Benchmark for Hierarchical Reinforcement Learning in Grid-Interactive Hyperscale Data Centers
NeurIPS 2026FusionInverse Modeling for Laser Pulse Shape Design in Inertial Confinement Fusion
NeurIPS 2026FusionEnd-to-End Differentiable Diffusion Conditioning for Physics-Informed Optimization
IJCAI 2026FusionHuman-in-the-Loop Meta Bayesian Optimization for Fusion Energy and Scientific Applications
AAAI 2026Data Center SurrogateFast 3D Surrogate Modeling for Data Center Thermal Management
NeurIPS 2025Data Center Liquid CoolingLC-Opt: Benchmarking Reinforcement Learning and Agentic AI for End-to-End Liquid Cooling Optimization in Data Centers
NeurIPS 2025Data Center CloudDCcluster-Opt: Benchmarking Dynamic Multi-Objective Optimization for Geo-Distributed Data Center Workloads
AAAI 2025Trustworthy AIReinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters
AAAI 2025Data CenterC CloudHierarchical Multi-Agent Framework for Carbon-Efficient Liquid-Cooled Data Center Clusters
NeurIPS 2024Data Center OptimizationSustainDC: Benchmarking for Sustainable Data Center Control
AAAI 2024Data Center OptimizationCarbon Footprint Reduction for Sustainable Data Centers in Real-Time
AAAI 2024Data Center OptimizationSustainability of Data Center Digital Twins with Reinforcement Learning
AAAI 2024Trustworthy AIRobustness and Visual Explanation for Black-box Image, Video, and ECG Signal Classification with Reinforcement Learning
WACV 2024Trustworthy AIBenchmark Generation Framework with Customizable Distortions for Image Classifier Robustness
IJCAI 2023Wave EnergyFunction Approximation for Reinforcement Learning Controller for Energy from Spread Waves
AAAI 2022Wave EnergyMulti-agent reinforcement learning controller to maximize energy efficiency for multi-generator industrial wave energy converter
Best Paper Awards
Selected recognitions and award-track papersNeurIPS 2023 WorkshopBest ML InnovationReal-time Carbon Footprint Minimization in Sustainable Data Centers with Reinforcement Learning
IEEE CASE 2022Best Application PaperSkip Training for Multi-Agent Reinforcement Learning Controller for Industrial Wave Energy Converters
ICML 2025 WorkshopBest Paper TrackInterpretable LLM Control for Sustainable Liquid Cooling in HPC Data Centers
Other Papers
Complete bibliography carried forward from the current site / public profile, grouped by year2026
AAAI 2026Thermal MLFast 3D Surrogate Modeling for Data Center Thermal Management
arXiv 2026Bayesian OptimizationBayMOTH: Bayesian optiMizatiOn with meTa-lookahead — a simple approacH
2025
ICML 2025 WorkshopBest Paper
Interpretable LLM Control for Sustainable Liquid Cooling in HPC Data Centers
NeurIPS 2025 WorkshopFusion
OptICF: Sample-Efficient Optimization of Implosion Outcomes in Inertial Confinement Fusion
NeurIPS 2025 WorkshopDiffusion / Fusion
DiffICF: Diffusion-Driven Inverse Modeling for Laser Pulse Design in Inertial Confinement Fusion
NeurIPS 2025 WorkshopLaser Pulse Design
Inverse Modeling of Laser Pulse Shapes in Inertial Confinement Fusion with Auto-Regressive Models
NeurIPS 2025 WorkshopAI for Science
IM-LPG: Inverse Modeling Approach to Laser Pulse Shape Generation in Inertial Confinement Fusion
NeurIPS 2025 WorkshopOptimization / Fusion
Fewer Shots, Better Implosions: Sample-Efficient Optimization for Inertial Confinement Fusion
NeurIPS 2025 WorkshopLLM Agents
Sustainable Control of Geo-Distributed Datacenters by Distilling Numerical Experts into Adaptive LLM Agents
NeurIPS 2025 WorkshopThermal Design
ML-Guided Cold Plate Design and Thermal Analysis for Liquid-Cooled HPC Servers
NeurIPS 2025 WorkshopRL / LLM Control
Carbon-Aware RL-LLM Control for Energy-Efficient Liquid-Cooled HPC Data Centers
NeurIPS 2025 WorkshopSustainable HPC
Efficient Reinforcement Learning Implementations for Sustainable Operation of Liquid Cooled HPC Data Centers
NeurIPS 2025 WorkshopGeo-Distributed DC
Holistic Sustainability for Geo-Distributed Data Centers using Hierarchical Optimization
CVPR 2025 WorkshopVideo Robustness
Robustness Evaluation for Video Models with Reinforcement Learning
CVPR 2025 WorkshopVLM Robustness
Coordinated Robustness Evaluation Framework for Vision-Language Models
NeurIPS 2025Data CentersDCcluster-Opt: Benchmarking Dynamic Multi-Objective Optimization for Geo-Distributed Data Center Workloads
NeurIPS 2025Liquid CoolingLC-Opt: Benchmarking Reinforcement Learning and Agentic AI for End-to-End Liquid Cooling Optimization in Data Centers
AAAI 2025Sustainable DCHierarchical Multi-Agent Framework for Carbon-Efficient Liquid-Cooled Data Center Clusters
AAAI 2025Adversarial RLReinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters
Workshop 2025Liquid CoolingLiquid Cooling Optimization for Data Centers with Reinforcement Learning
arXiv 2025LLM AgentsSurvey of LLM Agent Communication with MCP: A Software Design Pattern Centric Review
Preprint 2025LLM ControlA Survey on Large Language Models for Control Systems
Preprint 2025Self-CorrectionN-CRITICS: Self-Refinement of Large Language Models with Ensemble of Critics
2024
NeurIPS 2024 WorkshopRed Teaming
iART: Imitation Guided Automated Red Teaming
NeurIPS 2024 WorkshopLLM Reasoning
Informed Tree of Thought: Cost-Efficient Problem Solving with Large Language Models
NeurIPS 2024 WorkshopFusion / BO
LLM Enhanced Bayesian Optimization for Scientific Applications like Fusion
APS 2024Fusion
Optimizing the Performance of Direct-Drive Implosion Experiments Using Meta-Bayesian Optimization
NeurIPS 2024 WorkshopHuman Feedback
Explainable Meta Bayesian Optimization with Human Feedback for Scientific Applications like Fusion Energy
NeurIPS 2024 WorkshopFusion / BO
Meta-Learned Bayesian Optimization for Energy Yield in Inertial Confinement Fusion
IEEE Computer 2024Digital Twin
Digital Twins for Data Centers
NeurIPS 2024 WorkshopWave Energy
Enhancing Reinforcement Learning-Based Control of Wave Energy Converters Using Predictive Wave Modeling
NeurIPS 2024 WorkshopCloud / Carbon
Carbon-Aware Spatio-Temporal Workload Distribution in Cloud Data Center Clusters Using Reinforcement Learning
NeurIPS 2024 WorkshopLiquid Cooling
Enhancing Sustainability in Liquid-Cooled Data Centers with Reinforcement Learning Control
IEEE ITherm 2024CFD Surrogate
CFD Surrogates for Data Center Sustainability Using 3D U-Net Convolutional Neural Network
NeurIPS 2024 WorkshopSafe GenAI
Imitation Guided Automated Red Teaming
NeurIPS 2024BenchmarkSustainDC: Benchmarking for Sustainable Data Center Control
AAAI 2024Carbon-Aware RLCarbon Footprint Reduction for Sustainable Data Centers in Real-Time
AAAI 2024Digital TwinSustainability of Data Center Digital Twins with Reinforcement Learning
AAAI 2024RobustnessRobustness and Visual Explanation for Black-box Image, Video, and ECG Signal Classification with Reinforcement Learning
IEEE Computer 2024Digital TwinDigital Twins for Data Centers
Workshop 2024CFD SurrogateCFD Surrogates for Data Center Sustainability Using 3D U-Net Convolutional Neural Network
arXiv 2024BenchmarkSustainDC — Benchmarking for Sustainable Data Center Control
2023
NeurIPS Workshop 2023Best ML InnovationReal-Time Carbon Footprint Minimization in Sustainable Data Centers with Reinforcement Learning
arXiv 2023Sustainable DCPyDCM: Custom Data Center Models with Reinforcement Learning for Sustainability
NeurIPS Workshop 2023Sustainable DCSustainable Data Center Modeling: A Multi-Agent Reinforcement Learning Benchmark
NeurIPS Workshop 2023Digital TwinA Configurable Pythonic Data Center Model for Sustainable Cooling and ML Integration
NeurIPS Workshop 20233D CFDEnhancing Data Center Sustainability with a 3D CNN-Based CFD Surrogate Model
NeurIPS Workshop 2023Wave EnergyOcean Wave Energy: Optimizing Reinforcement Learning Agents for Effective Deployment
SafeAI @ AAAI 2023Adversarial MLRobustness with Black-Box Adversarial Attack using Reinforcement Learning
CVPRW 2023Adversarial MLRobustness with Query-Efficient Adversarial Attack using Reinforcement Learning
CVPRW 2023ExplainabilityRL-CAM: Visual Explanations for Convolutional Networks using Reinforcement Learning
CASE 2023Adversarial MLReinforcement Learning Based Black-Box Adversarial Attack for Robustness Improvement
CASE 2023Carbon-Aware RLConcurrent Carbon Footprint Reduction (C2FR) Reinforcement Learning Approach for Sustainable Data Center Digital Twin
CASE 2023Bayesian OptimizationRTDK-BO: High Dimensional Bayesian Optimization with Reinforced Transformer Deep Kernels
arXiv 2023RobustnessBenchmark Generation Framework with Customizable Distortions for Image Classifier Robustness
2022 and earlier
AAAI 2022Wave EnergySkip Training for Multi-Agent Reinforcement Learning Controller for Industrial Wave Energy Converters
IEEE CASE 2022Best Application PaperSkip Training for Multi-Agent Reinforcement Learning Controller for Industrial Wave Energy Converters
NeurIPS Workshop 2022ML SafetyMeasuring Robustness with Black-Box Adversarial Attack Using Reinforcement Learning
Earlier WorkTo CompleteAdditional earlier papers from the legacy site can be pasted here if the site contains older entries not exposed by the public search index.
Research Projects
Long-running themes and platformsExaDigiT and Liquid Cooling Digital Twins
High-fidelity simulation and AI-control environments for understanding and optimizing next-generation liquid-cooled supercomputing and data-center systems.
- Thermal forecasting and control
- Reinforcement learning environments
- Operator-facing explainability
AI-Energy Fabric
A research agenda for coordinating AI workloads, data centers, cloud platforms, cooling systems, and electric grids using auditable, safety-bounded AI.
- Grid-interactive data centers
- Resource-metabolic ML
- Bounded autonomy and governance
Trustworthy Agentic AI
Methods for evaluating, constraining, and improving LLM-based agents that reason over complex systems and take consequential actions.
- Stress testing and verification
- Reasoning quality and rule-following
- Human-in-the-loop oversight
AI for Fusion and Scientific Discovery
Machine learning for complex scientific systems, including surrogate modeling and inverse design for inertial confinement fusion workflows.
- Physics-informed surrogates
- Generative inverse design
- Scientist-in-the-loop optimization
Service & Collaboration
Academic, industrial, and national-lab collaborationsCommunity
Reviewer, organizer, contributor, and collaborator across machine learning, AI systems, trustworthy AI, sustainable computing, and AI for science communities.
Collaboration Interests
I welcome collaborations on ML for sustainable infrastructure, trustworthy agents, digital twins, high-assurance RL, and AI methods for scientific discovery.
