The simultaneous growth of new applications and death of transistor scaling trends is driving an explosion in custom silicon projects spanning all domains of computing. However, the enormous non-recurring engineering (NRE) cost of designing a modern system-on-a-chip (SoC) remains a major barrier to the wider adoption of custom silicon. Of concern to this talk is the lack of a good full-system...

I will talk about our recent works on computer vision: 1. Smart architecture search methods to discover a highly efficient model called EfficientNet, and EfficientNetV2 which are much more efficient than previous image classification models. 2. Using unlabeled data to improve accuracy and robustness of vision models. The method, named "Noisy Student", requires generating pseudo labels on...

In many important control applications the system of interest is high dimensional and is required to satisfy a complex specification. This talk focuses on addressing this challenge using hierarchical and distributed control techniques, with a focus on applications in transportation. In particular, we will explore the potential for connected automated vehicles (CAVs) to improve the safety and...

The field of linear control has seen broad application in fields as diverse as robotics, aviation, and power-grid maintenance. Accordingly, decades of research have been dedicated to learning to control systems when the dynamics are not known a priori, but must be inferred from data. However, much of the classical work considered asymptotic regimes, which shed little light on precisely how much...

The zoom location for this meeting is https://berkeley.zoom.us/j/99597287931?pwd=RjhQN1dEOTlrNG5PV20rbC93ejZ0Zz09 To mediate chromosome segregation during cell division, the microtubule-based spindle size adapts to changes in cell size and shape, which vary dramatically across species and within a multicellular organism. However, the nature of scaling events and their underlying mechanisms are...

Design of energy delivery systems is becoming an increasingly complex and multi-disciplinary field. In emerging applications such as renewable energy delivery, electric vehicle powertrains and charging, and grid interface systems, the power electronics are becoming more sophisticated and more densely-packed than ever before. In this talk, we will discuss considerations of such power electronics...

In this thesis, we pursue the design of Bayesian models as well as large-scale approximate inference procedures for the analysis of single-cell transcriptomics data. First, we rely on Bayesian statistics to formalize several central algorithmic steps of single-cell transcriptomics data analysis. In particular, we introduce single-cell Variational Inference (scVI), a deep generative model that...

Power electronics based ac-dc and dc-ac power conversion systems are critical links in the power delivery chains of today's electrical grid for emerging applications such as electric vehicle charging, renewable energy source integration data center power delivery, consumer electronics and so on. In this talk, the design and control techniques to improve the performance for both single-phase and...