Find Academic Supervisors

Stefano ErmonView profile

Associate Professor

• Machine Learning
• Generative AI
• Computational Sustainability
• +7 more

Stefano Ermon is an Associate Professor in the Department of Computer Science at Stanford University, affiliated with the Artificial Intelligence Laboratory and a Senior Fellow at the Woods Institute for the Environment. His research focuses on advancing machine learning and generative AI techniques to address societal and environmental challenges, including computational sustainability, geospatial analysis, and climate science. He holds a Ph.D. from Cornell University (2015). Education: Ph.D. in Computer Science, Cornell University (2015). Research Interests: Ermon’s work bridges foundational machine learning (e.g., diffusion models, generative AI, and optimization) with applications in sustainability, geospatial analysis (via satellite imagery), and earth observation systems. Notable contributions include predicting poverty using satellite data and developing scalable methods for molecule generation. Articles Trends: His recent work emphasizes diffusion models for generative tasks (e.g., text-to-image, molecule design), geospatial AI (e.g., environmental monitoring), and ethical AI (e.g., bias mitigation in LLMs). He also explores applications in robotics and scientific computing. Awards: He has received prestigious awards, including the ICML 2024 Best Paper Award, Sloan Research Fellowship, Microsoft Research Faculty Fellowship, and the IJCAI Computers and Thought Award. Advising and Grants: Ermon teaches courses like Probabilistic Graphical Models (CS228) and has secured grants from NSF, ONR, AFOSR, and private foundations. His lab develops tools for climate science and sustainable development. Labs/Teams: Leads the Stanford AI Lab group focused on computational sustainability and generative AI, collaborating with institutions like the Woods Institute for environmental applications.

• Computer Vision
• Machine Learning
• Visual Reasoning
• +7 more

Justin Johnson is an Assistant Professor at the University of Michigan's College of Engineering, Department of Electrical Engineering and Computer Science, and a Research Scientist at Facebook AI Research (FAIR). His work bridges computer vision, machine learning, and deep learning, focusing on visual reasoning, vision-language tasks, image generation, and 3D reasoning using neural networks. PhD, Stanford University (advised by Fei-Fei Li) His research interests span visual reasoning , vision and language , 3D vision , and image generation , with a focus on innovative applications of deep neural networks. Recent publications highlight work on 3D consistency, self-supervised learning, and multimodal integration of vision and text. Notable contributions include PyTorch3D for 3D data processing, and foundational work in visual question answering , neural style transfer , and scene graph-based image generation . Publications span top conferences like ICCV, CVPR, and NeurIPS. He teaches courses including EECS 498/598: Deep Learning for Computer Vision and EECS 442: Computer Vision at University of Michigan, with prior involvement in Stanford's CS 231N in co-teaching roles. Software projects include open-source frameworks like fast-neural-style for real-time artistic style transfer, and PyTorch3D for efficient 3D deep learning. These tools demonstrate his commitment to practical implementations and community-driven research.

Carl VondrickView profile

Professor

• Computer Vision
• Machine Learning
• Multimodal Perception
• +11 more

Carl Vondrick is a Professor in the Department of Computer Science at Columbia University. His research focuses on creating robust and versatile perception systems that leverage video and interaction with the natural world, with applications in 3D reconstruction, visual question answering, and robot manipulation. Former research scientist at Google Visiting researcher at Cruise Education: PhD (2017) from MIT, advised by Antonio Torralba BS (2011) from UC Irvine, advised by Deva Ramanan His research explores multimodal approaches for cross-task and cross-modal transfer, scene dynamics, audiovisual perception, interpretable models, and spatial awareness systems. The lab emphasizes zero-shot generalization and neuro-symbolic methods while addressing safety and robustness in AI systems. Key publication trends include: 2025: Video generation for robotics 2024: Differentiable rendering and cross-modal reasoning 2023: Robust perception and 3D modeling Scientific Awards: 2024 PAMI Young Researcher Award 2021 NSF CAREER Award Teaching Roles: Teaching Computer Vision II (2021-2025), Computer Vision I (2018-2019), and Representation Learning (2020-2022). Advising: Advises 8 current PhD students and has mentored 5 graduated students now at institutions like MBZUAI and UMD. The lab recruits 1-2 PhD students annually through Columbia’s PhD program. Grants and Collaborations: Funded by NSF, DARPA, Toyota Research Institute, Amazon Research, and Google.

Huan ZhangView profile

Assistant Professor

• Trustworthy Machine Learning
• Formal Verification of Machine Learning
• AI Safety
• +4 more

Huan Zhang serves as an Assistant Professor in the Department of Electrical and Computer Engineering at the University of Illinois Urbana-Champaign (UIUC), with affiliate appointments in the Department of Computer Science and the Coordinated Science Laboratory. His research focuses on building trustworthy AI systems with formal verification techniques to provide provable guarantees for safety-critical applications, particularly in machine learning and neural networks. Dr. Zhang received his Ph.D. in Computer Science from UCLA in 2020, advised by Professor Cho-Jui Hsieh. His academic journey includes an M.S. in Computer Engineering from UC Davis (2014) and a Bachelor of Engineering from Zhejiang University (2012). Prior to joining UIUC, he completed a postdoctoral fellowship at Carnegie Mellon University (2021-2023) with Professor Zico Kolter. Huan Zhang's research program centers on formal verification of machine learning systems, with particular emphasis on neural network verification, AI safety, robustness, and reliability. He pioneered the linear bound propagation-based verification framework that enables formal verification for networks with millions of neurons. His work spans five major research categories: formal verification of machine learning, training trustworthy ML models, machine learning safety and adversarial attacks, reinforcement learning safety, and optimization for scalable machine learning. His CROWN framework (NeurIPS 2018) established a foundational approach for neural network verification through efficient linear bound propagation. His recent publications demonstrate a strategic expansion from foundational verification techniques toward increasingly complex systems including large language models, vision-language models, and robotic control systems. The research trajectory shows a clear progression from theoretical frameworks to practical implementations with real-world applications, particularly in safety-critical domains. His work increasingly bridges formal methods with practical AI deployment requirements. Winner of International Verification of Neural Networks Competition (VNN-COMP) as team leader (2021-2024) Schmidt Futures AI2050 Early Career Fellowship ($300,000 research grant) Adversarial Machine Learning (AdvML) Rising Star Award (2021) IBM PhD Fellowship (2018) Dr. Zhang leads the development of α,β-CROWN, a neural network verifier that has won VNN-COMP 2021-2023, and auto_LiRPA, a PyTorch-based library for perturbation analysis on general computational graphs. He has mentored numerous graduate students from CMU, UCLA, UIUC, and Columbia University. His research is supported by significant funding including the Schmidt Futures fellowship and industry collaborations. He teaches courses including ECE 120, ECE 484, ECE 584, and ECE 598 HZ on topics ranging from computing fundamentals to safe autonomy and machine learning. Dr. Zhang maintains active research collaborations across multiple institutions and is affiliated with UIUC's Coordinated Science Laboratory. His work has significant implications for safety-critical AI applications in autonomous systems, healthcare, and other mission-critical domains where reliability guarantees are essential. He regularly gives guest lectures at institutions including Yale, Stony Brook, and the University of Nebraska Lincoln on formal verification techniques.

• Machine Learning
• Artificial Intelligence
• Neural Networks
• +3 more

Geoffrey E. Hinton is a distinguished Professor in the Department of Computer Science at the University of Toronto. He is renowned for his foundational contributions to machine learning, particularly in the development of deep learning and neural networks. His research focuses on understanding learning processes in both artificial and biological systems, with key contributions including Boltzmann machines, backpropagation, and deep belief networks. He teaches advanced machine learning courses such as CSC2535, emphasizing topics like graphical models, variational inference, and deep learning architectures. His work has been published extensively in top journals and conferences, with recent papers exploring forward-forward algorithms, analog diffusion models, and scalable neural network training methods. Hinton has advised numerous PhD and master's students and collaborates with institutions like Vector Institute. He is a central figure in the global AI community, regularly presenting at conferences (e.g., 2023 talks on CBS 60 Minutes, BBC, and PBS). His lab focuses on advancing machine learning theory and applications, addressing challenges in vision, language, and generative models.

Greg DurrettView profile

Associate Professor

• Natural Language Processing
• Machine Learning
• Artificial Intelligence
• +7 more

Greg Durrett is an Associate Professor in the Department of Computer Science at University of Texas at Austin, leading the TAUR Lab (Text Analysis, Understanding, and Reasoning ). His research focuses on advancing Large Language Models (LLMs) for knowledge-intensive tasks in medical information processing scientific discovery legal reasoning . He received his B.S. in Computer Science and Mathematics from MIT (2010) and Ph.D. in Computer Science from UC Berkeley (2016). His work develops techniques to train LLMs with new capabilities augment models for reliability assess model outputs improve reasoning frameworks . His 15 most recent publications (2021-2025) span knowledge propagation in LLMs chain-of-thought reasoning code generation benchmarks multi-modal reasoning fact verification discourse analysis . Scientific honors include NSF CAREER Award (2024) NSF grants (2018, 2024) Bloomberg Data Science Grant (2017) Facebook Fellowship (2014) Best Paper Finalist (EMNLP 2013) . Teaching: CS388: Natural Language Processing (graduate) CS371N: NLP (undergraduate) High school NLP module .

Alexander SchwingView profile

Associate Professor

• Machine Learning
• Computer Vision
• Structured Prediction
• +7 more

Alexander Schwing is an Associate Professor in the Department of Electrical and Computer Engineering and Computer Science at the University of Illinois at Urbana-Champaign, affiliated with the Coordinated Science Laboratory. His research focuses on machine learning and computer vision with applications in 3D scene understanding, generative modeling, and multi-agent systems. Education: Diploma in Electrical Engineering and Information Technology, Technical University of Munich (TUM) PhD in Computer Science, ETH Zurich Postdoctoral Fellow, University of Toronto Research Interests: Structured prediction in deep learning Generative adversarial networks and stability Multi-modal vision-language models 3D scene reconstruction from single images Embodied agent collaboration Semantic segmentation with temporal coherence Recent Publications: Highlight trends in neural rendering, video object segmentation, and reinforcement learning with applications to 3D modeling and multi-agent systems. Notable innovations include SAIL-VOS dataset for amodal segmentation and NeRFDeformer for single-view scene transformation. Scientific Awards: NSF CAREER Award, 3M and Amazon research awards, multiple student recognition awards, ETH Zurich PhD medal, and best paper at Intelligent Tutoring Systems 2014. Teaching: Offers graduate courses in Pattern Recognition (ECE 544) and Machine Learning (CS 446/ECE 449). Previously taught at University of Toronto and ETH Zurich. Labs & Collaborations: Leads research at Coordinated Science Laboratory (UIUC) with collaborations across University of Toronto, ETH Zurich, and industry partners like Samsung SAIT and Amazon.

Anthony RoweView profile

Professor

• Networked Embedded Systems
• Sensor Networks
• Extended Reality (XR) Systems
• +6 more

Anthony Rowe is the Siewiorek and Walker Family Professor of Electrical and Computer Engineering at Carnegie Mellon University (CMU) and a Chief Scientist at Bosch Research. His primary affiliation is with the CyLab and the Wireless, Sensing and Embedded Systems (WiSE Lab) at CMU. He specializes in networked embedded systems, sensor networks, and extended reality (XR) technologies. His research emphasizes energy-efficient sensing, real-time localization, and XR integration with physical systems. Research Focus: His work spans XR systems (e.g., AR/VR edge networking in ARENA), mmWave radar for sensing (e.g., tire wear monitoring via Osprey), distributed edge computing (Silverline), and low-power wide-area networking (OpenChirp). Recent efforts include AI-integrated XR platforms (XaiR) and radar tomography (DART). Grants & Projects: Leads the CONIX Research Center ($27.5M NSF/DARPA grant), Bosch-funded edge computing projects, and DOE initiatives on microgrids. Notable projects include ARENA (XR edge architecture), GridBallast (smart grid control), and rural microgrid deployments in Haiti. Awards: Best Student Paper (ISMAR 2024), Best Paper (IPSN 2020), and the Steven J. Fenves Research Award (2015). Recognized for innovations in localization (MobiCom 2021), radar (ICRA 2023), and energy systems (BuildSys 2010). Teaching: Teaches courses on embedded systems (18-349/18-449), real-time systems, and mixed reality (18-453). Courses emphasize hands-on design and real-world applications. Labs & Teams: Directs the WiSE Lab, collaborating with Bosch Research and industry partners. The lab develops open-source frameworks like ARENA and OpenChirp, and contributes to standards for edge computing and sensing.

Anca DraganView profile

Associate Professor

• Human-AI Interaction
• Human-Robot Interaction
• AI Safety and Alignment
• +5 more

Anca Dragan is an Associate Professor in the Department of Electrical Engineering and Computer Sciences at the University of California, Berkeley, where she runs the InterACT Lab focused on algorithms for human-AI and human-robot interaction. Currently on leave from Berkeley, she leads AI Safety and Alignment at Google DeepMind, overseeing safety for Gemini models and preparing for future advancements. Dragan has been a co-PI of the Center for Human-Compatible AI and served on the steering committee for the Berkeley AI Research (BAIR) Lab. B.Sc. in Computer Science from Jacobs University Bremen, Germany Ph.D. from Carnegie Mellon University Dr. Dragan's research focuses on enabling AI agents to work effectively with, around, and in support of people. Her work bridges robotics, machine learning, and game theory to create systems that better understand human preferences and coordinate with users. Key areas include AI alignment (ensuring AI does what people actually want), learning reward functions from diverse human feedback forms, and developing algorithms for human-AI collaboration across domains like autonomous vehicles, brain-machine interfaces, and recommender systems. Her research emphasizes maintaining uncertainty about human preferences and accounting for the plurality of human values. Dr. Dragan's recent publications reveal a strong focus on addressing fundamental challenges in AI safety and alignment. Her work spans theoretical foundations of reward learning, practical implementations for human-AI coordination, and critical examinations of limitations in current approaches. There's a clear trajectory toward more robust, safe, and value-aligned AI systems that can handle complex human preferences while avoiding both present-day harms and potential catastrophic risks. IEEE RAS Early Academic Career Award in Robotics and Automation (2021) McEntyre Award for Excellence in Teaching (2020) PECASE (Presidential Early Career Award for Science and Engineering) (2019) Sloan Fellowship (2018) NSF CAREER Award (2017) Okawa Foundation Award (2017) MIT Tech Review 35 Innovators Under 35 (2017) Multiple best paper awards at top robotics and AI conferences Dr. Dragan has mentored numerous successful students who have gone on to faculty positions at MIT, Stanford, CMU, and Princeton, as well as industry roles at DeepMind, Waymo, and Meta. Her advising philosophy emphasizes both technical rigor and consideration of broader societal impacts. She has secured significant research funding including NSF CAREER, ONR Young Investigator, and Okawa Foundation awards, supporting work on human-AI interaction and alignment. Dragan has also consulted for Waymo for six years, helping develop roadmaps for deploying increasingly learning-based safety-critical systems. Dr. Dragan leads the InterACT Lab at UC Berkeley, which has produced influential work on Cooperative Inverse Reinforcement Learning, Inverse Reward Design, and other foundational concepts in human-AI interaction. The lab's research has significantly shaped the field of AI alignment, with applications spanning autonomous vehicles that coordinate with human drivers, brain-machine interfaces that adapt to user needs, and language models that better understand human preferences. Current work focuses on scaling safety approaches as AI capabilities advance, ensuring alignment keeps pace with technological progress.

• Computer Vision
• Object Recognition
• Scene Understanding
• +2 more

Derek W Hoiem is a Professor in the Siebel School for Computing and Data Science at the University of Illinois Urbana-Champaign, where he has been a faculty member since 2009. His research focuses on computer vision and related areas, and he is also the co-founder and Chief Science Officer of Reconstruct, an AI-based construction technology company. His educational background includes: PhD in Robotics, Carnegie Mellon University (2007) Beckman Postdoctoral Fellowship (2008) Prof. Hoiem's research spans computer vision, with a focus on object recognition, scene understanding, and graphics. His work also extends to mobile robotics and 3D scene reconstruction. He has made significant contributions in areas such as visual recognition, 3D modeling, and the application of computer vision in construction monitoring. His recent publications (2023-2025) demonstrate a strong focus on advancing multimodal understanding, particularly in region-based representations, 3D vision, and neural radiance fields. There is a clear trend towards integrating language and vision, improving efficiency in neural networks, and applying computer vision to real-world problems such as construction progress monitoring. His scientific awards and honors are extensive and include: IEEE Fellow (2022) University Scholar (2022) Koendrink Prize (2022) Dean's Award for Excellence in Research, Associate Professor (2021) Campus Distinguished Promotion Award (2015) Best Paper Award: IEEE Winter Conference on Applications in Computer Vision (WACV) (2015) CW Gear Junior Faculty Award (2014) IEEE PAMI Young Researcher Award (2014) Dean's Award for Excellence in Research, Assistant Professor (2014) Sloan Research Fellowship (2013) Intel Early Career Faculty Honor Program Award (2012) NSF CAREER Award (2011) ACM Doctoral Dissertation Award, Honorable Mention (2008) Carnegie Mellon University SCS Distinguished Dissertation Award (2008) Best Paper Award: IEEE Computer Vision and Pattern Recognition (CVPR) (2006) Prof. Hoiem has secured significant research funding, including an NSF CAREER award and an Intel Early Career Faculty award. He is also actively involved in technology transfer, having co-founded Reconstruct where he serves as Chief Science Officer. His teaching excellence is reflected in multiple "List of Teachers Ranked as Excellent" awards spanning from 2010 to 2021. Prof. Hoiem leads a research group at UIUC focused on computer vision and 3D scene understanding. Additionally, he co-founded and serves as Chief Science Officer at Reconstruct, which develops AI-based solutions for construction monitoring.

Noah D. GoodmanView profile

Associate Professor

• Computational models of cognition
• Probabilistic programming languages
• Natural language semantics and pragmatics
• +6 more

Noah D. Goodman is Associate Professor of Psychology and Computer Science, and Linguistics (by courtesy) at Stanford University. He directs the Computation & Cognition Lab (CoCoLab) at Stanford, where he leads research on computational models of cognition, integrating logic and probability. His work spans cognitive psychology, linguistics, and computer science. Primary Appointment: Psychology Department By Courtesy: Computer Science Department and Linguistics Department Director: Computation & Cognition Lab (CoCoLab) Goodman's research focuses on computational models of cognition, with particular interest in probabilistic approaches to understanding human thought. His work integrates logic and probability to model concepts, categorization, intuitive theories, causal learning and reasoning, social cognition (including reasoning about others' goals, beliefs, and actions), cognitive development (especially acquisition of abstract knowledge), and natural language semantics and pragmatics. He has made significant contributions to the development of probabilistic programming languages as tools for cognitive modeling. His recent publications demonstrate a strong trend toward integrating probabilistic modeling with linguistic theory and social cognition. The articles span computational cognitive science, natural language processing, and artificial intelligence, with a consistent theme of using probabilistic frameworks to understand complex cognitive phenomena. Many papers explore how humans make inferences under uncertainty across different domains. Goodman teaches several courses at Stanford including Language and Thought (Psych 132), Computation and Cognition: the Probabilistic Approach (Psych 204/CS 428), Foundations of Cognition (Psych 205), and Introduction to Cognitive Science. He has also led seminars on topics ranging from natural and artificial intelligence to the science of meditation.

Eric WongView profile

Assistant Professor

• Machine Learning
• Artificial Intelligence
• Explainable AI
• +5 more

Eric Wong is an Assistant Professor in the Department of Computer and Information Science at the University of Pennsylvania, affiliated with the ASSET Center and leading the Brachio Lab. His research focuses on robust and reliable machine learning, including model debugging, adversarial robustness, and explainable AI. He holds a PhD from Carnegie Mellon University (CMU), advised by J. Zico Kolter, and completed a postdoc with Aleksander Madry. His work bridges theory and practice, addressing challenges in model interpretability, safety, and scalability. Teaching includes CIS 5200 (Machine Learning), CIS 3333 (Mathematics for Machine Learning), and a specialized course on debugging ML pipelines. Notable contributions include the FIX Benchmark for interpretable features and defenses against LLM jailbreaking attacks. He received an Amazon Research Award in 2024 and has published extensively in top conferences like ICML, NeurIPS, and ICLR. Affiliations: University of Pennsylvania (CIS), ASSET Center, Brachio Lab Education: PhD in Machine Learning (CMU), Postdoc at MIT Key Projects: FIX Benchmark, DOLPHIN framework, SmoothLLM defense Awards: Amazon Research Award (2024) Lab Focus: Safe AI, model debugging, neurosymbolic learning

Bahar AsgariView profile

Assistant Professor

• Computer Architecture
• Domain-Specific Architecture Design
• Near Memory Processing
• +3 more

Bahar Asgari is an Assistant Professor in the Department of Computer Science at the University of Maryland, College Park. She holds appointments in CS, UMIACS, and ECE, and directs the Computer Architecture and Systems Lab (CASL). Her research focuses on domain-specific architecture design, near memory processing, and reconfigurable computing. Before joining UMD, she worked at Google, contributing to system optimization and memory utilization strategies. Education: Ph.D., Georgia Institute of Technology, 2021. Research Interests: Developing energy-efficient architectures for sparse problems, scientific computing acceleration, FPGA optimizations, and reconfigurable hardware systems. Her work addresses challenges in memory efficiency, parallelization, and hardware-software co-design across domains like machine learning and IoT. Key Awards: 2024 UMD Excellence in Teaching Award 2023 DOE Early Career Award Advising & Grants: Supervises 9 graduate students and leads projects sponsored by NSF, DOE, and industry partnerships. Her lab explores systolic arrays, near-data processing, and fault-tolerant computing. Labs/Teams: Directs CASL, which builds next-generation computing systems through hardware innovation and algorithm-architecture co-design.

Alois Christian KnollView profile

Professor

• Robotics
• Artificial Intelligence
• Autonomous Systems
• +6 more

Prof. Alois Christian Knoll is a full professor at the Technical University of Munich (TUM) in the School of Computation, Information and Technology. His academic career includes roles at Bielefeld University and leadership in major EU initiatives like the Human Brain Project and ECHORD++. He specializes in robotics, AI, and autonomous systems, with a focus on medical robotics, sensor-based systems, and neuromorphic engineering. Knoll has supervised over 100 doctoral theses and authored/co-authored over 1,000 publications. Education: Diploma in Electrical Engineering (University of Stuttgart, 1985); PhD in Computer Science (Technical University of Berlin, 1988); Habilitation (TU Berlin, 1993). He has been at TUM since 2001, leading the Robotics, AI, and Real-Time Systems department. Research interests span autonomous systems, neuro-IT integration, and traffic simulation. Key projects include fortiss (Bavarian State Institute for Computer Science) and TUM-CREATE (Singapore collaboration). Awards include IEEE Fellow, University of Tokyo Fellow, and the Carl-Ramsauer-Prize (1990). Current roles include editorships in robotics journals, leadership in EU flagship projects, and teaching across multiple programs. His work bridges computer science, neuroscience, and engineering, with applications in healthcare, automotive systems, and urban mobility.

David DuvenaudView profile

Associate Professor

• Machine Learning
• Generative Models
• Bayesian Inference
• +5 more

David Duvenaud is an Associate Professor at the University of Toronto , holding a Canada Research Chair in Generative Models and a Schwartz Reisman Chair in Technology and Society . He is cross-appointed to the Department of Computer Science and Department of Statistical Sciences . A Sloan Research Fellow and founding member of the Vector Institute , his work bridges deep probabilistic models , AI safety , and scientific computing . PhD in Machine Learning (University of Cambridge, 2014) Postdoc in Hyperparameter Optimization (Harvard University, 2016) Co-founded Invenia (energy forecasting company) His research spans foundational Neural Ordinary Differential Equations (NeurIPS 2018 Best Paper) and Automatic Chemical Design (ACS Central Science 2018) to recent work on AGI governance (2025) and AI safety (2024). Key contributions include stochastic variational inference , implicit differentiation frameworks , and antisymmetrization layers for quantum Monte Carlo. Recent publications (2024-2025) focus on systemic existential risks from AI , many-shot jailbreaking attacks , and epistemic uncertainty quantification . His group trains energy-based models with scalable MCMC samplers and develops invertible neural architectures (e.g., Residual Flows NeurIPS 2019). He also explores human-AI alignment through LLM Processes (NeurIPS 2024) and Sycophancy in Language Models (ICLR 2024). Canada Research Chair (2025) NSERC Grant (2025) Sloan Research Fellow (2021) Schwartz Reisman Chair (2021) Best Paper Award (NeurIPS 2018) Distinguished Paper Award (ICFP 2021) His students include James Requeima , Jesse Bettencourt , and Raymond Douglas . He teaches courses on Statistical Methods for Machine Learning and Differentiable Inference . Current work (2025) investigates systemic human disempowerment through incremental AI capabilities and sabotage risk mitigation via hyperparameter-aware evaluations.