Machine Learning

Data-driven approaches to design intelligent algorithms.

MERL has a long history of research activity in machine learning, including the development of various boosting algorithms and contributing to the theory and practice of highly scalable collaborative filtering. Our recent work has focused on deep learning and reinforcement learning, with application to a wide range of applications including automotive, robotics, factory automation, transportation, as well as building and home systems.

Quick Links
Researchers
Awards
- AWARD MERL team wins the Generative Data Augmentation of Room Acoustics (GenDARA) 2025 Challenge
  Date: April 7, 2025
  Awarded to: Christopher Ick, Gordon Wichern, Yoshiki Masuyama, François G. Germain, and Jonathan Le Roux
  MERL Contacts: Jonathan Le Roux; Yoshiki Masuyama; Gordon Wichern
  Research Areas: Artificial Intelligence, Machine Learning, Speech & Audio
  Brief
  - MERL's Speech & Audio team ranked 1st out of 3 teams in the Generative Data Augmentation of Room Acoustics (GenDARA) 2025 Challenge, which focused on “generating room impulse responses (RIRs) to supplement a small set of measured examples and using the augmented data to train speaker distance estimation (SDE) models". The team was led by MERL intern Christopher Ick, and also included Gordon Wichern, Yoshiki Masuyama, François G. Germain, and Jonathan Le Roux.
    
    The GenDARA Challenge was organized as part of the Generative Data Augmentation (GenDA) workshop at the 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2025), and held on April 7, 2025 in Hyderabad, India. Yoshiki Masuyama presented the team's method, "Data Augmentation Using Neural Acoustic Fields With Retrieval-Augmented Pre-training".
    
    The GenDARA challenge aims to promote the use of generative AI to synthesize RIRs from limited room data, as collecting or simulating RIR datasets at scale remains a significant challenge due to high costs and trade-offs between accuracy and computational efficiency. The challenge asked participants to first develop RIR generation systems capable of expanding a sparse set of labeled room impulse responses by generating RIRs at new source–receiver positions. They were then tasked with using this augmented dataset to train speaker distance estimation systems. Ranking was determined by the overall performance on the downstream SDE task. MERL’s approach to the GenDARA challenge centered on a geometry-aware neural acoustic field model that was first pre-trained on a large external RIR dataset to learn generalizable mappings from 3D room geometry to room impulse responses. For each challenge room, the model was then adapted or fine-tuned using the small number of provided RIRs, enabling high-fidelity generation of RIRs at unseen source–receiver locations. These augmented RIR sets were subsequently used to train the SDE system, improving speaker distance estimation by providing richer and more diverse acoustic training data.
- AWARD Mitsubishi Electric Team Wins Awards at GalFer Contest
  Date: June 23, 2025
  Awarded to: Bingnan Wang, Tatsuya Yamamoto, Yusuke Sakamoto, Siyuan Sun, Toshiaki Koike-Akino, and Ye Wang
  MERL Contacts: Toshiaki Koike-Akino; Bingnan Wang; Ye Wang
  Research Areas: Machine Learning, Multi-Physical Modeling, Optimization
  Brief
  - The MELSUR (Mitsubishi Electric SURrogate) team, consisting of a group of MERL and Mitsubishi Electric researchers, ranked first in two out of three categories in the GalFer Contest.
    
    The GalFer (Galileo Ferraris) contest aims to compare the accuracy and efficiency of data-driven methodologies for the multi-physics simulation of traction electric machines. A total of 26 teams worldwide participated in the contest, which consists of three categories. The MELSUR team, including MERL staff Bingnan Wang, Toshiaki Koike-Akino, Ye Wang, MERL intern Siyuan Sun, Mitsubishi Electric researchers Tatsuya Yamamoto and Yusuke Sakamoto, ranked first for the category of "Novelty" and "Interpolation". The results were announced during an award ceremony at the COMPUMAG 2025 conference in Naples, Italy.
- AWARD MERL work receives IEEE Transactions on Automation Science and Engineering Best New Application Paper Award from IEEE Robotics and Automation Society
  Date: May 19, 2025
  Awarded to: Yehan Ma, Yebin Wang, Stefano Di Cairano, Toshiaki Koike-Akino, Jianlin Guo, Philip Orlik, Xinping Guan and Chenyang Lu
  MERL Contacts: Stefano Di Cairano; Jianlin Guo; Toshiaki Koike-Akino; Philip V. Orlik; Yebin Wang
  Research Areas: Communications, Control, Machine Learning
  Brief
  - The paper “Smart Actuation for End-Edge Industrial Control Systems”, co-authored by MERL intern Yehan Ma, MERL researchers Yebin Wang, Stefano Di Cairano, Toshiaki Koike-Akino, Jianlin Guo, and Philip Orlik, and academic collaborators Xinping Guan and Chenyang Lu, was recognized as the Best New Application Paper of the IEEE Transactions on Automation Science and Engineering (T-ASE), for "a new industrial automation solution that ensures safety operation through coordinated co-design of edge model predictive control and local actuation".
    
    The award recognizes the best application paper published in T-ASE over the previous calendar year, for the significance of new applications, technical merit, originality, potential impact on the field, and clarity of presentation.
See All Awards for Machine Learning
News & Events
- TALK [MERL Seminar Series 2026] Laixi Shi presents talk titled Robust Decision Making Without Compromising Learning Efficiency
  Date & Time: Wednesday, January 14, 2026; 1:00 PM
  Speaker: Laixi Shi, Johns Hopkins University
  MERL Host: Dehong Liu
  Research Areas: Artificial Intelligence, Control, Machine Learning
  Abstract
  - Decision-making artificial intelligence (AI) has revolutionized human life ranging from healthcare, daily life, to scientific discovery. However, current AI systems often lack reliability and are highly vulnerable to small changes in complex, interactive, and dynamic environments. My research focuses on achieving both reliability and learning efficiency simultaneously when building AI solutions. These two goals seem conflicting, as enhancing robustness against variability often leads to more complex problems that requires more data and computational resources, at the cost of learning efficiency. But does it have to?
    
    In this talk, I overview my work on building reliable decision-making AI without sacrificing learning efficiency, offering insights into effective optimization problem design for reliable AI. To begin, I will focus on reinforcement learning (RL) — a key framework for sequential decision-making, and demonstrate how distributional robustness can be achieved provably without paying statistical premium (additional training data cost) compared to non-robust counterparts. Next, shifting to decision-making in strategic multi-agent systems, I will demonstrate that incorporating realistic risk preferences—a key feature of human decision-making—enables computational tractability, a benefit not present in traditional models. Finally, I will present a vision for building reliable, learning-efficient AI solutions for human-centered applications, though agentic and multi-agentic AI systems.
- NEWS MERL Researcher Diego Romeres Collaborates with Mitsubishi Electric and University of Padua to Advance Physics-Embedded AI for Predictive Equipment Maintenance
  Date: December 10, 2025
  MERL Contact: Diego Romeres
  Research Areas: Artificial Intelligence, Machine Learning, Robotics
  Brief
  - Mitsubishi Electric Research Laboratories (MERL) researchers, together with collaborators at Mitsubishi Electric’s Information Technology R&D Center in Kamakura, Kanagawa Prefecture, Japan, and the Department of Information Engineering at the University of Padua, have developed a cutting-edge physics-embedded AI technology that substantially improves the accuracy of equipment degradation estimation using minimal training data. This collaborative effort has culminated in a press release by Mitsubishi Electric Corporation announcing the new AI technology as part of its Neuro-Physical AI initiative under the Maisart program.
    
    The interdisciplinary team, including MERL Senior Principal Research Scientist and Team Leader Diego Romeres and University of Padua researchers Alberto Dalla Libera and Giulio Giacomuzzo, combined expertise in machine learning, physical modeling, and real-world industrial systems to embed physics-based models directly into AI frameworks. By training AI with theoretical physical laws and real operational data, the resulting system delivers reliable degradation estimates on the torque of robotic arms even with limited datasets. This result addresses key challenges in preventive maintenance for complex manufacturing environments and supports reduced downtime, maintained quality, and lower lifecycle costs.
    
    The successful integration of these foundational research efforts into Mitsubishi Electric’s business-scale AI solutions exemplifies MERL’s commitment to translating fundamental innovation into real-world impact.
See All News & Events for Machine Learning
Research Highlights
Internships
See All Internships for Machine Learning
Openings
- MS0268: Research Scientist - Multiphysical Systems
- CI0177: Postdoctoral Research Fellow - Agentic AI
See All Openings at MERL
Recent Publications
- Khattar, V., Choudhury, M., Rashid, M.R.U., Liu, J., Koike-Akino, T., Jin, M., Wang, Y., "Amplification Effects in Test-Time Reinforcement Learning: Safety and Reasoning Vulnerabilities", AAAI Workshop on Trust and Control in Agentic AI, January 2026.
  BibTeX TR2026-020 PDF
  - @inproceedings{Khattar2026jan,
  - author = {Khattar, Vanshaj and Choudhury, Moumita and Rashid, Md Rafi Ur and Liu, Jing and Koike-Akino, Toshiaki and Jin, Ming and Wang, Ye},
  - title = {{Amplification Effects in Test-Time Reinforcement Learning: Safety and Reasoning Vulnerabilities}},
  - booktitle = {AAAI Workshop on Trust and Control in Agentic AI},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-020}
  - }
- Koike-Akino, T., Chen, X., Liu, J., Wang, Y., Wang, P., Brand, M., "LatentLLM: Activation-Aware Transform to Multi-Head Latent Attention", AAAI Conference on Artificial Intelligence, January 2026.
  BibTeX TR2026-018 PDF Presentation
  - @inproceedings{Koike-Akino2026jan,
  - author = {{{Koike-Akino, Toshiaki and Chen, Xiangyu and Liu, Jing and Wang, Ye and Wang, Pu and Brand, Matthew}}},
  - title = {{{LatentLLM: Activation-Aware Transform to Multi-Head Latent Attention}}},
  - booktitle = {AAAI Conference on Artificial Intelligence},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-018}
  - }
- Rashid, M.R.U., Dasu, V.A., Wang, Y., Tan, G.G., Mehnaz, S., "Chain-of-Thought Driven Adversarial Scenario Extrapolation for Robust Language Models", AAAI Conference on Artificial Intelligence, January 2026.
  BibTeX TR2026-017 PDF
  - @inproceedings{Rashid2026jan,
  - author = {Rashid, Md Rafi Ur and Dasu, Vishnu Asutosh and Wang, Ye and Tan, G. Gary and Mehnaz, Shagufta},
  - title = {{Chain-of-Thought Driven Adversarial Scenario Extrapolation for Robust Language Models}},
  - booktitle = {AAAI Conference on Artificial Intelligence},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-017}
  - }
- Yataka, R., Wang, P., Boufounos, P.T., Takahashi, R., "Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion", AAAI Conference on Artificial Intelligence, January 2026.
  BibTeX TR2026-019 PDF Software
  - @inproceedings{Yataka2026jan,
  - author = {Yataka, Ryoma and Wang, Pu and Boufounos, Petros T. and Takahashi, Ryuhei},
  - title = {{Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion}},
  - booktitle = {AAAI Conference on Artificial Intelligence},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-019}
  - }
- Jeon, C.-B., Wichern, G., Germain, F.G., Le Roux, J., "Embracing Cacophony: Explaining and Improving Random Mixing in Music Source Separation", IEEE Open Journal of Signal Processing, January 2026.
  BibTeX TR2026-012 PDF
  - @article{Jeon2026jan,
  - author = {Jeon, Chang-Bin and Wichern, Gordon and Germain, François G and {Le Roux}, Jonathan},
  - title = {{Embracing Cacophony: Explaining and Improving Random Mixing in Music Source Separation}},
  - journal = {IEEE Open Journal of Signal Processing},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-012}
  - }
- Wilkinghoff, K., Yang, H., Ebbers, J., Germain, F.G., Wichern, G., Le Roux, J., "Local Density-Based Anomaly Score Normalization for Domain Generalization", IEEE Transactions on Audio, Speech and Language Processing, January 2026.
  BibTeX TR2026-010 PDF Software
  - @article{Wilkinghoff2026jan,
  - author = {Wilkinghoff, Kevin and Yang, Haici and Ebbers, Janek and Germain, François G and Wichern, Gordon and {Le Roux}, Jonathan},
  - title = {{Local Density-Based Anomaly Score Normalization for Domain Generalization}},
  - journal = {IEEE Transactions on Audio, Speech and Language Processing},
  - year = 2026,
  - month = jan,
  - url = {https://www.merl.com/publications/TR2026-010}
  - }
- Cornell, S., Boeddeker, C., Park, T., Huang, H., Raj, D., Wiesner, M., Masuyama, Y., Chang, X., Wang, Z.-Q., Squartini, S., Garcia, P., Watanabe, S., "Recent Trends in Distant Conversational Speech Recognition: A Review of CHiME-7 and 8 DASR Challenges", Computer Speech & Language, DOI: 10.1016/j.csl.2025.101901, Vol. 97, pp. 101901, December 2025.
  BibTeX TR2026-008 PDF
  - @article{Cornell2025dec,
  - author = {Cornell, Samuele and Boeddeker, Christoph and Park, Taejin and Huang, He and Raj, Desh and Wiesner, Matthew and Masuyama, Yoshiki and Chang, Xuankai and Wang, Zhong-Qiu and Squartini, Stefano and Garcia, Paola and Watanabe, Shinji},
  - title = {{Recent Trends in Distant Conversational Speech Recognition: A Review of CHiME-7 and 8 DASR Challenges}},
  - journal = {Computer Speech \& Language},
  - year = 2025,
  - volume = 97,
  - pages = 101901,
  - month = dec,
  - doi = {10.1016/j.csl.2025.101901},
  - url = {https://www.merl.com/publications/TR2026-008}
  - }
- Masuyama, Y., Wichern, G., Germain, F.G., Ick, C., Le Roux, J., "SuDaField: Subject- and Dataset-Aware Neural Field for HRTF Modeling", IEEE Open Journal of Signal Processing, DOI: 10.1109/OJSP.2025.3627073, Vol. 6, pp. 1169-1178, December 2025.
  BibTeX TR2026-009 PDF
  - @article{Masuyama2025dec2,
  - author = {Masuyama, Yoshiki and Wichern, Gordon and Germain, François G and Ick, Christopher and {Le Roux}, Jonathan},
  - title = {{SuDaField: Subject- and Dataset-Aware Neural Field for HRTF Modeling}},
  - journal = {IEEE Open Journal of Signal Processing},
  - year = 2025,
  - volume = 6,
  - pages = {1169--1178},
  - month = dec,
  - doi = {10.1109/OJSP.2025.3627073},
  - url = {https://www.merl.com/publications/TR2026-009}
  - }
See All Publications for Machine Learning
Videos

[MERL Seminar Series Spring 2026] Robust Decision Making Without Compromising Learning Efficiency

[ICASSP 2025] Indoor Airflow Imaging Using Physics-Informed Schlieren Tomography

[ICCV 2025] SAC-GNC: SAmple Consensus for adaptive Graduated Non-Convexity

[BMVC 2025] Towards Open-Vocabulary Multimodal 3D Object Detection with Attributes

[ICCV Workshop 2025] Joint Training of Image Generator and Detector for Road Defect Detection

[ICCV 2025] Toward Long-Tailed Online Anomaly Detection through Class-Agnostic Concepts

[CVPR 2025] TailedCore: Few-Shot Sampling for Unsupervised Long-Tail Noisy Anomaly Detection

FDPP: Fine-tune Diffusion Policy with Human Preference

[MERL Seminar Series Spring 2025] Red Teaming AI Agents in-the-wild: Revealing Deployment Vulnerabilities

[MERL Seminar Series Spring 2025] The Emergence of Generalizability and Semantic Low-Dim Subspaces in Diffusion Models

[MERL Seminar Series Spring 2025] Amplifying human performance in combinatorial competitive programming

[WACV 2025] Towards Zero-shot 3D Anomaly Localization

Data-driven Spatial Classification using Multi-Arm Bandits for Monitoring with Robot Teams

[NeurIPS 2024] MEL-PETs Defense for the NeurIPS 2024 LLM Privacy Challenge Blue Team Track

[NeurIPS 2024] MEL-PETs Joint-Context Attack for the NeurIPS 2024 LLM Privacy Challenge Red Team Track

[NeurIPS 2024] Evaluating Large Vision-and-Language Models on Children's Mathematical Olympiads

[MERL Seminar Series Fall 2024] Audio for Object and Spatial Awareness

[IROS 2024] Few-shot Transparent Instance Segmentation for Bin Picking

[MERL Seminar Series Fall 2024] Tools from cognitive science to understand the behavior of large language models

[ECCV 2024] PS-NEUS: A Probability-guided Sampler for Neural Implicit Surface Rendering

[ECCV 2024] Equivariant Spatio-Temporal Self-Supervision for LiDAR Object Detection

[CVPR 2024] SIRA: Scalable Inter-frame Relation and Association for Radar Perception

Decentralized, Safe, Multi-agent Motion Planning for Drones Under Uncertainty via Filtered Reinforcement Learning

[MERL Seminar Series Spring 2024] Neural Certificates and LLMs in Large-Scale Autonomy Design

[MERL Seminar Series Spring 2024] Decoding Hidden Worlds: Unprecedented Sensing and Connectivity for Climate, Robotics, & Smart Environments

Gear-NeRF: Free-Viewpoint Rendering and Tracking with Motion-aware Spatio-Temporal Sampling

[MERL Seminar Series Spring 2024] Are Emergent Abilities of Large Language Models a Mirage?

MERL's Quantum AI Technology

[GLOBECOM 2022/VCC 2023 Tutorial] Post-Deep Learning Era: Emerging Quantum Machine Learning for Sensing and Communications (Session 2)

[GLOBECOM 2022/VCC 2023 Tutorial] Post-Deep Learning Era: Emerging Quantum Machine Learning for Sensing and Communications (Session I)

[MERL Seminar Series Fall 2023] Robust and Physics-informed machine learning for low light imaging

[MERL Seminar Series Fall 2023] Multiplicity in Machine Learning

Multi-level Reasoning for Robotic Assembly: From Sequence Inference to Contact Selection

[MERL Seminar Series Fall 2023] Visual Programming - A compositional approach to building General Purpose Vision Systems

[MERL Seminar Series Fall 2023] The Confluence of Vision, Language, and Robotics

[MERL Seminar Series Fall 2023] A Process Systems Engineering Perspective on Carbon Capture: Key Challenges and Opportunities

[TRO 2024] Safe Multiagent Motion Planning Under Uncertainty for Drones Using Filtered Reinforcement Learning

Are Deep Neural Networks SMARTer than Second Graders?

[CVPR 2023] EVAL: Explainable Video Anomaly Localization

[MERL Seminar Series Spring 2023] Learning and Dynamical Systems

[MERL Seminar Series Spring 2023] Investigating Multi-Agent Reinforcement Learning for Grid-Interactive Smart Communities using CityLearn

[MERL Seminar Series Spring 2023] Pitfalls and Opportunities in Interpretable Machine Learning

[MERL Seminar Series Spring 2023] Neural Implicit Flow

[MERL Seminar Series Spring 2023] Towards Complex Language in Partially Observed Environments

[MERL Seminar Series Spring 2022] Hybrid robotics and implicit learning

Toshiaki Koike-Akino Gives Seminar Talk at IEEE Boston Photonics

[MERL Seminar Series Spring 2022] RLMPC: An Ideal Combination of Formal Optimal Control and Reinforcement Learning?

[MERL Seminar Series Spring 2022] Self-Supervised Scene Representation Learning

[MERL Seminar Series Spring 2022] Learning Speech Representations with Multimodal Self-Supervision

[MERL Seminar Series Spring 2022] Extreme optics design as a large-scale optimization problem

HealthCam: A system for non-contact monitoring of vital signs

[MERL Seminar Series 2021] Harnessing machine learning to build better Earth system models for climate projection

[MERL Seminar Series 2021] Deep probabilistic regression

[MERL Seminar Series 2021] Learning to See by Moving: Self-supervising 3D scene representations for perception, control, and visual reasoning

Control of Mechanical Systems via Feedback Linearization Based on Black-Box Gaussian Process Models

Application of Deep Learning for Nanophotonic Device Design (Invited)

Towards Human-Level Learning of Complex Physical Puzzles

Tactile-RL for Insertion: Generalization to Objects of Unknown Geometry

Scene-Aware Interaction Technology

Action Detection Using A Deep Recurrent Neural Network

MERL Research on Autonomous Vehicles

Semantic Scene Labeling

Obstacle Detection

Deep Hierarchical Parsing for Semantic Segmentation

Global Local Face Upsampling Network
Software & Data Downloads

Software & Data Downloads

MERL is making Machine Learning software and data available to the research community:

multi-view Radar object dEtection with 3D bounding boX diffusiOn (REXO)
Radar-based 3D Pose Estimation using Transformer (RAPTR)
MMHOI Dataset: Modeling Complex 3D Multi-Human Multi-Object Interactions (MMHOI)
Subject- and Dataset-Aware Neural Field for HRTF Modeling (SuDaField)
Open Vocabulary Attribute Detection Dataset (OVAD)
Task-Aware Unified Source Separation (TUSS)

See All Downloads