Dissertation Talk: Interpretable Machine Learning with Applications in Neuroscience

540AB DOP Center Cory Hall
  • Reza Abbasi Asl, UC Berkeley, Department of EECS
In the last decade, research in machine learning has been exceedingly focused on the development of algorithms and models with remarkably high predictive capabilities. Models such as convolutional neural networks (CNNs) have achieved state-of-the-art predictive performance for many tasks in computer vision, autonomous driving, and transfer learning in areas such as computational neuroscience....

Optogenetic and Chemogenetic Tools for Mapping Molecular and Cellular Circuits

60 Evans Hall
  • Prof. Alice Ting, Stanford University, Genetics/Biology/Chemistry
The first part of the talk will describe optogenetic tools we have developed for labeling and manipulating functional circuits in the brain (e.g., FLARE and related tools). The second part of the talk will describe chemogenetic tools we have developed for mapping molecular interactions in living cells (e.g., APEX and TurboID). ******** Alice Ting did her PhD in Chem here at UC Berkeley...

Dissertation Talk: Image Synthesis for Self-Supervised Learning

250 Sutardja Dai Hall
  • Richard Zhang, UC Berkeley, Department of EECS
We explore the use of deep networks for image synthesis, both as a graphics goal and as an effective method for representation learning. We propose BicycleGAN, a general system for image-to-image translation problems, with the specific aim of capturing the multimodal nature of the output space. We study image colorization in greater detail and develop automatic and user-guided approaches....
Title: Physically-based Modeling and Rendering of Complex Visual Appearance Abstract: Photorealistic Rendering in Computer Graphics is increasingly important. Whether in movies or video games, breathtaking graphics have become one of the most crucial factors to their success. However, state of the art rendering still struggles with two fundamental challenges -- realism and speed. It...

50 Shades of Streaming with DASH

405 Soda Hall
  • Ali C. Begen, Ozyegin University, Turkey
HTTP adaptive streaming (HAS) has become the technology of choice for delivering video content over the Internet. We first show that when multiple HAS clients share a network bottleneck, bad things happen. To address the problems, we present a number of solution approaches. Second, we discuss the ways to achieve consistent-quality streaming. Third, we look into some of the recent developments in...

Dissertation Talk: Next Generation Datacenter Architecture

420 Soda Hall
  • Peter Xiang Gao
Modern datacenters are the foundation of large scale Internet services, such as search engines, cloud computing and social networks. In this talk, I will investigate the new challenges in building and managing large scale datacenters. Specifically, I will show trends and challenges in the software stack, the hardware stack and the network stack of modern datacenters, and propose new approaches to...

Dissertation Talk: Avoiding Communication in First-Order Methods for Optimization

306 (HP Auditorium) Soda Hall
  • Aditya Devarakonda
Iterative ML methods are particularly sensitive to communication cost since they often require communication every iteration. We derive communication-avoiding variants of BCD methods that solve popular regression, classification, and related kernel problems. We show that these variants are numerically stable and can attain large speedups of up to 6.1x on a Cray XC30 supercomputer.

Dissertation Talk: Exploratory model analysis for machine learning

405 Soda Hall
  • Biye Jiang, UC Berkeley, Department of EECS
Machine learning is growing in importance in many different fields. However, it is still very hard for users to tune hyper-parameters when optimizing their models, or perform a comprehensive and interpretable diagnosis for complex models like deep neural nets. Existing developer tool like TensorBoard only provides limited functionality which usually visualizes model statistics based on metrics...
Due to limited platform size, most portable and wearable electronics call for a compact power management achitecture with high efficiency and high power density. The conventional buck converter requires bulky magnetics that consume a large amount of board areas. The switched capacitor (SC) converter, though promising for a fully integrated solution, is not efficient for continuous voltage...

Reframing Tech Speaker Series with Dr. Marcia Linn

306 Soda Hall
  • Dr. Marcia Linn, UC Berkeley Graduate School of Education
Bias Busters is hosting the Reframing Tech Speaker Series in Spring & Fall 2018 to start a discussion about the issues underlying the lack of representation of people of color, women, people with disabilities, and other minorities in STEM, as well as wider ethical issues in tech. Each talk will be followed by a discussion about actionable steps we can all take – as students, faculty, staff;...