Pu Wang

About Pu Wang

Passionate about building radio-frequency (RF) and multimodal systems that enable models to sense, perceive, understand, reason, and act from multi-sensor data; Experienced in leading end-to-end deep learning (DL) and signal processing (SP) developments—spanning data collection, pipeline design, neural network modeling, evaluation, and delivery of demos and prototypes.40+ publications at Trans, with contributions to open-source repos and datasets and industry standards (e.g, 802.11bf Wi-Fi sensing).Email: p••••••••@gmail.com; p••••@merl.comGoogle Scholar: & hl=enGitHub: https://github.com/pwang-oxford (during my Oxford visit)https://github.com/pwang_merlPersonal webpage (@MERL):

Experience

1. Senior Principal Research Scientist

   Mitsubishi Electric Research Laboratories

   Apr 2016 — Present · Cambridge, MA, US

   1) LLM, LMRM (Large Multi-Modal Reasoning Models), and Sensor Reasoning Models* Efficient inference, Multi-modal fusion, Sensor reasoning* LatentLLM Activation-Aware Transform to Multi-Head Latent Attention* TuneComp (CVPR\'25 WS): Joint Fine-Tuning and Compression for Large Foundation Models* Slaying the HyDRA (NeurIPS\'24 WS): Parameter-Efficient Hyper Networks with Low-Displacement Rank Adaptation* SuperLoRA (BMVC\'24): Parameter-Efficient Unified Adaptation of Large Foundation Models* LoDA (NeurIPS\'23 ENLSP WS): Low-Dimensional Adaptation of Large Language Models* ABRNet (IJCAI\'22): Adversarial Bi-Regressor Network for Domain Adaptive Regression* IMVDA (ECCV\'22): Multi-sensor domain adaptation 2) Radar/RF Perception and Generation (Initiator, Project Lead, and Individual Contributor)* Deep learning for radar perception (object detection, pose estimation, segmentation, occupancy prediction) and generation (neuro-physical simulator, scene generation)* REXO Indoor Multi-View Radar Object Detection via 3D Bounding Box Diffusion* RAPTR (NeurIPS\'25): Radar-based 3D Pose Estimation using Transformer* RETR (NeurIPS\'24): Multi-View Radar Detection Transformer for Indoor Perception (Code: https://github.com/merlresearch/radar-detection-transformer)* MMVR (ECCV\'24): Millimeter-wave Multi-View Radar Dataset and Benchmark for Indoor Perception (Dataset: https://zenodo.org/records/12••••78)* SIRA (CVPR\'24): Scalable Inter-frame Relation and Association for Radar Perception * TempoRadar (CVPR\'22): Exploiting Temporal Relations on Radar Perception for Autonomous Driving 3) Wireless Sensing (Initiator, Project Lead, and Individual Contributor)* Deep learning for Wi-Fi/UWB/5G/BLE localization, sensing, interference mitigation.* 60-GHz/Multi-Band neural dynamic learning JSTSP\'24 & TWC\'25)* Mutual Interference Mitigation for MIMO-FMCW Automotive Radar TVT\'24)* Technical contributions to 802.11bf WLAN Sensing Task Group

Education

• Stevens Institute of Technology

  Doctor of Philosophy (Ph.D.), Electrical Engineering

  2007 — 2011