Bichen Wu

Title: Bichen Wu: Innovator in Neural Network Design

Introduction

Bichen Wu is a prominent inventor based in Menlo Park, CA (US). She has made significant contributions to the field of neural network design, particularly through her innovative patent. Her work focuses on enhancing the efficiency of neural networks, which is crucial in the rapidly evolving landscape of artificial intelligence.

Latest Patents

Bichen Wu holds a patent titled "Hardware-aware efficient neural network design system having differentiable neural architecture search." This patent describes computer-implemented systems that utilize a differentiable neural architecture search (DNAS) engine. The DNAS engine operates on one or more processors and is configured with a stochastic super net that defines a layer-wise search space. This search space includes a variety of candidate layers, each specifying one or more operators for a neural network architecture. The DNAS engine processes training data to train weights for these operators based on a loss function that represents the latency of each operator on a target platform. It selects a set of candidate neural network architectures from the trained stochastic super net, optimizing the architecture distribution through gradient-based methods.

Career Highlights

Bichen Wu is currently employed at Meta Platforms, Inc., where she continues to push the boundaries of neural network technology. Her work is instrumental in developing systems that are not only efficient but also adaptable to various hardware platforms.

Collaborations

Bichen collaborates with talented individuals such as Peizhao Zhang and Peter Vajda, contributing to a dynamic and innovative work environment.

Conclusion

Bichen Wu's contributions to neural network design exemplify the intersection of technology and innovation. Her patent reflects a commitment to advancing artificial intelligence through efficient and effective neural network architectures.