Expanding Applications for ML Through Research
As machine learning (ML) expands to more applications across all areas of compute and the wider technology agenda, our research continues to guide and inform this growth. Arm advanced hardware, software, and tools provide the energy efficiency and performance required to support increasingly complex algorithms in this rapidly evolving area.
Arm Research Impact Report 2021
Intelligent applications have the potential to improve billions of lives. Find out how in the Arm Research Impact Report 2021.
Key Research Threads
Our research covers a wide range of topics that focus on developing the technology to power future machine learning solutions.
Podcast: AI in Business
|Efficient Residue Number System Based Winograd Convolution||Zhi-Gang Liu, Matthew Mattina
|Systolic Tensor Array: An Efficient Structured-Sparse GEMM Accelerator for Mobile CNN Inference
||Zhi-Gang Liu, Paul Whatmough , Matthew Mattina
|Run-Time Efficient RNN Compression for Inference on Edge Devices||Urmish Thakker, Jesse Beu, Dibakar Gope, Ganesh Dasika, Matthew Mattina|
|Compressing RNNs for IoT Devices by 15-38x Using Kronecker Products||Urmish Thakker, Jesse Beu, Dibakar Gope, Chu Zhou, Igor Fedorov, Ganesh Dasika, Matthew Mattina
|Sparse Architecture Search for CNNs on Resource-Constrained Microcontrollers
||Igor Fedorov, Ryan P. Adams, Matthew Mattina, Paul N. Whatmough
|Urmish Thakker, Ganesh Dasika, Jesse Beu, Matthew Mattina
|Ternary Hybrid Neural-Tree Networks for Highly Constrained IoT Applications
||Dibakar Gope, Ganesh Dasika, Matthew Mattina
|Efficient Winograd or Cook-Toom Convolution Kernel Implementation on Widely Used Mobile CPUs
||Partha Maji, Andrew Mundy, Ganesh Dasika, Jesse Beu, Matthew Mattina, Robert Mullins
|Learning Low-Precision Neural Networks without Straight-Through Estimator(STE)||Zhi-Gang Liu, Matthew Mattina
|RNN Compression using Hybrid Matrix Decomposition
||Urmish Thaker, Jesse Beu, Dibakar Gope, Ganesh Dasika, Matthew Mattina
|DNN Engine: A 28-nm Timing-Error Tolerant Sparse Deep Neural Network Processor for IoT Applications
||Paul Whatmough, S.K. Lee, D. Brooks, G.Y. Wei
|FixyNN: Efficient Hardware for Mobile Computer Vision via Transfer Learning
||Paul N. Whatmough, Chuteng Zhou, Patrick Hansen, Shreyas Kolala Venkataramanaiah, Jae-sun Seo, Matthew Mattina
|Efficient and Robust Machine Learning for Real-World Systems
||Franz Pernkopf, Wolfgang Roth, Matthias Zoehrer, Lukas Pfeifenberger, Guenther Schindler, Holger Froening,Sebastian Tschiatschek, Robert Peharz, Matthew Mattina, Zoubin Ghahramani
|Energy Efficient Hardware for On-Device CNN Inference via Transfer Learning
||Paul Whatmough, Chuteng Zhou, Patrick Hansen, Matthew Mattina
|SCALE-Sim: Systolic CNN Accelerator Simulator
||Ananda Samajdar, Yuhao Zhu, Paul Whatmough, Matthew Mattina, Tushar Krishna
|Euphrates: Algorithm-SoC Co-Design for Low-Power Mobile Continuous Vision
||Yuhao Zhu, Anand Samajdar, Matthew Mattina, Paul Whatmough
|Mobile Machine Learning Hardware at ARM: A Systems-on-Chip (SoC) Perspective||Yuhao Zhu, Matthew Mattina, Paul Whatmough
Meet the Team
Senior Director of Machine Learning
Matthew Mattina is head of Arm’s Machine Learning Research Lab, where he leads a team of world-class researchers developing advanced hardware, software, and algorithms for machine learning.
Join the team! We are always looking for talented researchers across all areas of ML. In particular, we are keen to hear from experts in probabilistic ML, including Bayesian inference, Gaussian processes, variational inference, probabilistic models, and ensemble learning. See our current vacancies.
Latest ML Research Blogs
Read more blogs on our community website.
SpArSe: Democratizing and Enabling TinyML on Arm M-class
Microcontrollers (MCUs) are truly the ubiquitous computer of our time. They are tiny, cheap, and use low power. They can often be powered indefinitely using a solar cell. They are in your watch, your fridge, and your car contains about 30.
Reducing the Cost of Neural Network Inference with Residue Number Systems
With neural network model size and computational complexity continuing to grow exponentially, the increase in computational requirements presents major adoption challenges. Could Residue Number Systems be the solution?
Adapting Models to the Real World: On-Device Training for Edge Model Adaptation
Neural networks are becoming widely used in computer interaction, but in real-world scenarios we see errors. Our edge distillation research could hold the answer to increased accuracy and efficiency.