On-device automatic speech recognition is now becoming feasible and is useful in scenarios without internet connection or when data privacy is a concern.
FixyFPGA: Fully-parallel and fully-pipelined FPGA accelerator for sparse CNNs
Most conventional FPGA-based accelerators use off-chip memory for data transformation, then perform computation for a single-layer in a time-multiplexed manner. Throughput is often limited by the memory…
Neural network architectures for deploying TinyML applications on commodity microcontrollers
TinyML seeks to deploy ML algorithms on ultra low power systems, to enable us to intelligently select which data to transmit, improving energy efficiency.
Improving federated learning with dynamic regularization
IoT devices collect and transmit data to the cloud, where it is then analyzed and ML models trained. Privacy challenges arise when users are reluctant to share their personal data.
What you missed at the second On-Device Intelligence Workshop
The workshop held in conjunction with MLSys 2021 brought researchers and practitioners together to discuss key issues, share new research results and practical tutorial material.
Arm Research’s collaboration with the Cambridge ELLIS Unit
Arm Research is excited to be part of the Cambridge ELLIS Unit, focusing on two key elements: Bayesian statistics and Probabilistic ML.
Ensuring your AI is sure: Any place, anywhere, anytime
It is important in industry to define what we see and how well we see it. This simple yet powerful idea has driven recent developments in the Arm Research ML Lab.
Using multiple labels improves neural network learning
A single label is not enough. Label diversity can be introduced by creating several labels for each training example in a way that the ordinal structure allows.
Research for a sustainable future
To help companies find the breakthrough innovations needed to support the Global Goals, the UNGC set up the Young SDG Innovator Program, which our colleagues in Arm joined.
Efficient Bug Discovery with Machine Learning for Hardware Verification
For present-day microprocessors, it is even more challenging to identify bugs. Using (ML) to efficiently identify bugs, we've seen a 25% increase in efficiency than the default verification workflow. …
Reducing the Cost of Neural Network Inference with Residue Number Systems
The size and computational complexity of neural network models continues to grow exponentially. However, the increase in computational requirements presents a major challenge to their adoption. Could Residue…
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. We’ve recently completed research into edge distillation to solve this problem.
It is time for natively flexible processors
The story behind our flexible processors paper started with how to make billions of everyday things smart.
Scalable Hyperparameter Tuning for AutoML
Mango is an open source Python library for hyperparameter optimization, built for AutoML systems. Developed by Arm Research, Mango presents many useful features.
Even Faster Convolutions: Winograd Convolutions meet Integer Quantization and Architecture Search
The design of deep learning (DL) neural network (NN) models targeting mobile devices has advanced rapidly over the last couple of years. Important computer vision tasks have led a community-wide transition…
SCALE-Sim: A cycle-accurate NPU simulator for your research experiments
Architecture simulators are a key tool in the computer architecture toolbox. They provide a convenient model of real hardware at a level of abstraction that makes them faster and more flexible than low…
TinyML Applications Require New Network Architectures
Researchers have studied neural network compression for quite some time. However, the need for always on compute has led to a recent trend towards executing these applications on even smaller IoT devices…
A Year of Discovery: Arm Research 2019
2019 was yet another year of incredible technology discovery for Arm Research. Inspiring advancements have been made across the research community, and Arm Research has contributed to this. Our teams have…