Dates
Conferences
Tags
Sort by:
Most recent
Conference: KubeCon + CloudNativeCon Europe 2023
Authors: Chad Torbin
2023-04-20
Join us for the world premiere of Inside Envoy, a captivating documentary that delves into the origins and rapid ascent of one of the most significant open source projects in the community today. This groundbreaking film will transport you to the forefront of the action, where you'll witness firsthand how the project emerged as an in-house solution within ride-share giant Lyft, before rapidly evolving into an innovation that has defined the careers of those who helped create a proxy that fundamentally transformed the industry. You'll follow the journeys of many of the most talented engineers in the field as they recount the story like never before, providing a behind-the-scenes look at the remarkable rise of this industry-changing project. Watch the "Inside Envoy - The Proxy for the Future" movie trailer: https://www.youtube.com/watch?v=sQVeuFYvzIk
Conference: Transform X 2022
Authors: Dr. Craig Martell, Alexandr Wang
2022-10-19
tldr - powered by Generative AI
The speaker discusses the importance of getting data right for successful implementation of AI in the Department of Defense. They also highlight the need for integrating cutting-edge technologists into the defense industrial base and the potential of AI in logistics.
- AI is happening massively across the Department of Defense, but getting the data right allows for building a Marketplace that does AI correctly
- Integrating cutting-edge technologists into the defense industrial base is crucial for successful implementation of AI
- Reinforcement learning techniques can be applied to logistics challenges in the Department of Defense for massive taxpayer impact
Conference: Transform X 2022
Authors: Laura Major
2022-10-19
tldr - powered by Generative AI
Motional's approach to developing autonomous vehicles involves continuous learning and data sharing across the industry.
- Motional uses a continuous learning framework to mine on-road driving data and discover rare scenarios or areas where there are challenges or issues with their performance.
- They up sample and incorporate more of these scenarios into their training data to improve their autonomy performance.
- Motional recognizes the need for richer development of data sets and sharing of those data sets to fuel the development across the industry.
- They have pioneered a data sharing culture that has now extended across the industry.
- Motional's approach involves not just increasing the volume of data, but getting the right data, including finding rare objects and identifying challenging scenarios.
- Their focus is on improving their autonomy performance to achieve true driverless capability.
Conference: Transform X 2022
Authors: John List
2022-10-19
tldr - powered by Generative AI
The presentation discusses the complementarities between AI and behavioral economics, with examples from the speaker's experiences as Chief Economist at Uber, Lyft, and Walmart. The speaker emphasizes the importance of effect identification, architectural nudges, and detection of heterogeneity in using AI to improve the bottom line and make the world better.
- The speaker shares examples of using AI and behavioral economics to improve products and scaling
- Effect identification, architectural nudges, and detection of heterogeneity are important tools in using AI to improve the bottom line and make the world better
- Examples include the economics of apologies at Uber, pricing and wait times at Lyft, and the voltage effect on scaling
- The speaker emphasizes the importance of combining economic theory, field experiments, and AI to identify causal effects and correlations
Conference: OpenAI + Data Forum 2022
Authors: Jakub Piotr Cłapa, Marcus Edel
2022-06-23
Data sets are the backbone of Machine-Learning (ML), but some are more critical than others. There is a core set of them that researchers use to evaluate machine-learning models as a way to track how ML capabilities are advancing over time. One of the most known is the ImageNet data set, which kicked off the modern ML revolution. There's also Lyft's data set meant to train self-driving cars, etc. Over the years, studies have found that these data sets can contain serious flaws. ImageNet, for example, has several labels that are just flat-out wrong. A mushroom is labeled a spoon, a lion is labeled a monkey, or in the case of the Lyft data set, several cars are not annotated at all. All these datasets have one thing in common; they use a highly error-prone annotation pipeline with little or no quality checks. We worked on an open-source tool that uses and combines novel unsupervised machine-learning pipelines that help annotators and machine-learning engineers to identify and filter out potential label errors. In this talk, we will share our findings on how label errors affect the existing training process, discuss possible implications, and dive into how we leveraged unsupervised learning to filter out annotation errors while looking at real-world examples.