Troubleshoot Deep Learning Models / Simple multi-user Jupyter Notebooks

San Francisco PyData
San Francisco PyData
Public group
Location image of event venue


The SF PyData meetup group is kicking off our 2019 season with a bang!

1. Jupyter notebooks have become an essential tool for all types of data analyses. Jupyter project contributor Yuvi Panda will teach us the simplest way to jump straight into using and sharing Jupyter notebooks, without the pain of wrestling with environments and setup.

2. Deep learning continues to surprise and delight with its ability to solve thorny problems and achieve impressive results. But neural networks aren't exactly known for their transparency and ease-of-use. OpenAI research scientist (and UC Berkeley phD student) Josh Tobin will teach us the simplest way to debug misbehaving deep learning models and get your analysis back on track.

As always, we'll also have time for you to meet, mingle, and discuss all things data with other Bay Area Python and data enthusiasts!

## Agenda:

6:00 - 6:45pm: Mingling
6:45 - 6:50pm: Opening remarks
6:50 - 7:35pm: Tech-Talk-1: Simple multi-user Jupyter Notebooks with The Littlest JupyterHub
7:35 - 8:05pm: Tech-Talk-2: How to Troubleshoot Your Deep Learning Models
8:05 - 8:30pm: Mingling
8:30: Event over!

Many thanks to Cloudflare for volunteering to host.

Doors will close at 7:15 so please arrive before then.

## Simple multi-user Jupyter Notebooks with "The Littlest JupyterHub"

#### Abstract
JupyterHub is a multi-user notebook server, allowing any number of users to access their own Jupyter-based data science environment without having to download or install anything. This is extremely useful when teaching classes or conducting trainings - you can skip straight to your content without having to get people to install Anaconda or mess around in a cloud console. Analytics teams also benefit from this, since they can easily access the data they are analyzing and share work between themselves. The Littlest JupyterHub ( is a JupyterHub distribution that makes it easy to setup and maintain a JupyterHub with minimal long term effort. This talk will demo a quick setup, and various use cases for a small JupyterHub.

#### Speaker Bio
Yuvi Panda is a contributor to Project Jupyter, and currently works at UC Berkeley building & running JupyterHubs for their data science courses. He also is part of the team that runs the open infrastructure behind He primarily works on removing accidental complexities from people's workflows so they can focus on the tasks they wanna do rather than fiddle with unnecessary 'computer stuff'. Prior to this he was part of the Wikimedia community, working on enabling Wikimedia volunteers to build bots and run analysis as easily as possible.

## How to Troubleshoot Your Deep Learning Models
#### Abstract:
Deep learning practitioners spend most of their time troubleshooting & debugging. Troubleshooting models is notoriously difficult because the same performance problem can be attributed to many different sources, and performance can be extremely sensitive to small changes in architecture and hyperparameters. In this talk, I will attempt to demystify the troubleshooting process by presenting a decision tree for improving your model's performance.

#### Speaker Bio:
Josh Tobin is a Research Scientist at OpenAI and a PhD student in Computer Science at UC Berkeley working with Professor Pieter Abbeel. Josh's research focuses on applying deep learning to problems in robotic perception and control, with a particular concentration on robotic manipulation, deep reinforcement learning, generative models, and synthetic data. Prior to Berkeley and OpenAI, Josh was a management consultant at McKinsey & Co. in New York. Josh has a BA in Mathematics from Columbia University.