Small is Beautiful: Building Compact DNNs with Reinforcement Learning
With the emergence of Deep Neural Networks (DNNs) on mobile and embedded devices, DNNs must satisfy strict computing constraints, which often limits performance and prevents people from using AI on these devices. Deeplite researches and develops intelligent optimization software powered by Reinforcement Learning (RL) to automatically design robust, real-time & low-power DNN models. We'll take a look at the current state of automated DNN model design, the challenges with RL approaches, and how such solutions will make AI more accessible.
- Deep learning algorithms deliver incredible performance, but require massive amounts of time, compute and expertise to implement.
- Deeplite helps humans automatically design deep learning models that are optimized for their task.
- the future is small. smaller, more efficient deep learning will enable new use cases and business value to be created.
Davis is Co-founder & Product Lead at Deeplite, a Montreal-based AI startup. He leads a team leveraging years of research from the Brown University SCALE Lab, USC and Deeplite on new developments in reinforcement learning to provide flexible and powerful deep learning optimization software for industry. Davis also works with industry leaders in consumer, automotive and IoT markets to find new ways to make deep learning more affordable. Prior to Deeplite, Davis co-founded a fintech startup and developed statistical models for drug safety at Takeda Oncology.