AI Efficiency Researcher · Qualcomm AI Research

Roohollah
Amiri

Model Quantization & Neural Network Compression —
Building AI that fits on your phone

I work on making large AI models — LLMs, vision networks, perception pipelines — run fast and accurately on mobile and edge hardware, without sacrificing quality. Based at Qualcomm AI Research, San Diego.

Publications Get in Touch
Scholar GitHub LinkedIn DBLP
Roohollah Amiri

Roohollah Amiri

Staff AI Researcher @ Qualcomm
430+
Citations
10+
Publications
5+
Yrs Industry
PhD
Boise State
Quantization Edge AI 5G / 6G RF-SLAM RL
01 · About

Who I Am

I am a Staff AI Researcher at Qualcomm AI Research in San Diego, focused on AI efficiency — compressing and quantizing large neural networks so they run fast and accurately on-device, from smartphones to autonomous vehicles. My core work spans post-training quantization, mixed-precision compression, and making LLMs and vision models viable on Qualcomm's Snapdragon hardware without sacrificing quality.

Before moving to AI Research, I spent four years on Qualcomm's Wireless R&D team applying deep learning to 5G/6G problems: building neural RF-SLAM systems for GPS-free indoor positioning, constructing digital twin networks, and developing AI-native localization pipelines — including a live demo at MWC 2024 in Barcelona.

I hold a Ph.D. in Electrical and Computer Engineering from Boise State University (dissertation: Reinforcement Learning in Self-Organizing Cellular Networks, advised by Prof. Hani Mehrpouyan), and was a visiting scholar at the University of Texas at Austin collaborating with Prof. Jeffrey Andrews.

430+ Citations
10+ Publications
5+ Years at Qualcomm
Model Quantization On-Device AI LLM Compression Edge Inference 5G / 6G RF-SLAM Reinforcement Learning Autonomous Systems
Roohollah Amiri
Position Staff AI Researcher
Employer Qualcomm AI Research
Location San Diego, CA
Email amiriruhollah@gmail.com
Google Scholar UovQ3w0AAAAJ
ORCID 0000-0002-6373-121X
02 · Experience

Career

Staff AI Researcher
Qualcomm AI Research · San Diego, CA
Aug 2024 – Present

Research on model quantization and compression for efficient on-device AI inference, with a focus on deploying Mixture-of-Experts (MoE) models across the full hardware spectrum.

  • On-device deployment of MoE models ranging from 4B to 1T+ parameters — spanning mobile devices, automotive platforms, and cloud inference
  • Post-training and quantization-aware training of LLMs and vision models
  • Mixed-precision quantization targeting Snapdragon SoC (Hexagon NPU, Adreno GPU)
  • AI model optimization for smartphones, autonomous vehicles, and XR devices
  • Contributing to AIMET — Qualcomm's open-source model efficiency toolkit
Senior Engineer — Wireless R&D
Qualcomm Technologies, Inc. · San Diego, CA
2020 – Aug 2024

Applied deep learning to 5G/6G positioning, mapping, and network intelligence.

  • Neural RF-SLAM: unsupervised indoor positioning and environment mapping from 5G CSI
  • AI-native 5G localization with IMU supervision — sub-meter accuracy (GLOBECOM 2023)
  • Digital twin construction and validation on live over-the-air testbed — showcased at MWC 2024
  • Indoor environment learning via RF-mapping, published in IEEE JSAC 2023
Visiting Research Scholar
University of Texas at Austin · Austin, TX
2019 – 2020

Collaboration with Prof. Jeffrey Andrews on large-scale 5G HetNet simulation and reinforcement learning for network optimization.

  • Reinforcement learning for power optimization in neural network-driven heterogeneous networks
  • Developed GeoNS — open-source geometric network simulator with spatial indexing for large-scale 5G evaluation
PhD Research Assistant
Boise State University · Boise, ID
2016 – 2020

Dissertation: Reinforcement Learning in Self-Organizing Cellular Networks — Advisor: Prof. Hani Mehrpouyan.

  • Distributed multi-agent RL for power control in 5G heterogeneous networks
  • Machine learning for mmWave MIMO and self-organizing small cells
  • 9 publications; key paper on ML for HetNets cited 200+ times
03 · Publications

Selected Publications

2024
IEEE Access
Advancing Next Generation Wireless Networks With Digital Twin: Construction, Validation, and Real-World Applications on an Indoor Over-the-Air Testbed
Akgun B., Jolly A., Sachdev B., Ravichandran D., Amiri R., et al.
IEEE Xplore
2024
IEEE VTC Fall
Digital Twin Powered Next Generation Wireless Networks: Construction, Validation, and Applications
Jolly A., Akgun B., Sachdev B., Ravichandran D., Jayabalan M., Amiri R., et al.
IEEE Xplore
2023
IEEE JSAC
Indoor Environment Learning via RF-Mapping
Amiri R., Yerramalli S., Yoo T., Hirzallah M., Zorgui M., Prakash R., Zhang X.
IEEE Xplore
2023
GLOBECOM
Neural 5G Indoor Localization with IMU Supervision
Ermolov A., Kadambi S., Arnold M., Hirzallah M., Amiri R., et al.
IEEE Xplore
2022
ICC
Neural RF SLAM for Unsupervised Positioning and Mapping with Channel State Information
Kadambi S., Behboodi A., Soriaga J.B., Welling M., Amiri R., Yerramalli S., Yoo T.
IEEE Xplore
2020
IEEE TWC
mmWave Lens-Based MIMO System for Suppressing Small-Scale Fading and Shadowing
Almasi M.A., Amiri R., Jafarkhani H., Mehrpouyan H.
IEEE Xplore
2020
IEEE VTC
Spatial Indexing for System-Level Evaluation of 5G Heterogeneous Cellular Networks
Amiri R., Balevi E., Andrews J.G., Mehrpouyan H.
IEEE Xplore
2019
IEEE TWC
Reinforcement Learning for Self Organization and Power Control of Two-Tier Heterogeneous Networks
Amiri R., Almasi M.A., Andrews J.G., Mehrpouyan H.
IEEE Xplore
2018
ICC
A Machine Learning Approach for Power Allocation in HetNets Considering QoS
Amiri R., Mehrpouyan H., Fridman L., Maleh R., Wicks M., Matolak D.
IEEE Xplore
All Publications on Google Scholar
04 · Projects

Research & Open Source

⚡
On-Device AI Efficiency
Post-training and quantization-aware compression of LLMs and vision models for Snapdragon SoC deployment. INT8/INT4 at production quality — no retraining required.
Quantization LLMs Edge AI Snapdragon
AIMET on GitHub
🌐
Digital Twin Network
High-fidelity virtual replica of a live 5G over-the-air testbed. Trains AI policies in simulation and transfers them to real hardware. Showcased live at MWC 2024.
Digital Twin 5G AI-Native MWC 2024
Watch Demo
📡
Neural RF-SLAM
Simultaneous positioning and environment mapping from 5G channel state information. No GPS, no prior maps — learns space geometry purely from radio signal propagation.
RF-SLAM 5G CSI Positioning Unsupervised
🔬
GeoNS — Geometric Network Simulator
Open-source C++ simulator for large-scale 5G networks using spatial indexing (k-d trees). Supports parallel experiments, distributed RL, and realistic heterogeneous network evaluation.
Open Source C++ 5G Sim Spatial DB
View on GitHub
🤖
Reinforcement Learning for Self-Organizing Networks
Distributed multi-agent reinforcement learning for autonomous power control and topology management in 5G HetNets — small cells that coordinate without a central controller.
Multi-Agent RL HetNets mmWave Self-Org
PhD Dissertation
05 · Education

Academic Background

Ph.D.
Electrical & Computer Engineering
Boise State University
2016 – 2020 · Boise, ID
Dissertation: Reinforcement Learning in Self-Organizing Cellular Networks. Advisor: Prof. Hani Mehrpouyan.
M.Sc.
Electrical Engineering
Iran University of Science & Technology
2014 – 2016 · Tehran, Iran
B.Sc.
Electrical Engineering
Iran University of Science & Technology
2010 – 2014
06 · Contact

Get in Touch

amiriruhollah@gmail.com LinkedIn Google Scholar GitHub ORCID DBLP