Autonomous Vehicles Gain Tracking Accuracy, Meta announces Velox, Improved Multi-GPU Support for Notebooks with Lightning and more!

Researchers are improving the workflows of genomic analysis and the tracking ability of autonomous vehicles. Experts warn of a proposal to regulate general-purpose AI in the EU, the Go programming language adds vulnerability management support and Meta announces Velox. Let’s dive in!

Research Highlights

• Researchers from Northeastern University in China proposed a Point-Track-Transformer (PTT) module for point cloud-based 3D single object tracking tasks for robotics and autonomous driving. Current approaches struggle with tracking objects at a long distance because of sparse or partially-occluded point clouds, making the features extracted by the model ambiguous, leading to poor tracking results. The proposed model generates fine-tuned attention features by computing attention weights to improve tracking ability in complex scenarios and surpasses the baseline by ~10% in the autonomous vehicle category.
• Genomepy , a freely available project released by researchers from Radboud University in the Netherlands, is attempting to improve the way genomic researchers analyze data and run bioinformatic workflows. Current standard practice includes tedious and error-prone steps like manual data inputting and searching through a plethora of providers for the right reference data. Genomepy automates the searching, downloading, and preprocessing of the right genomic data for analysis.
• Motivated by the success of Transformers in sequence modeling tasks, researchers from the University of Geneva in Switzerland introduced IRIS to solve the data inefficiencies commonly found within deep reinforcement learning agents. IRIS is an agent that learns in a world model composed of a discrete auto-encoder and an autoregressive Transformer– claiming to surpass MuZero by outperforming humans on 10 out of 26 games.

ML Engineering Highlights

• OpenAI introduced “outpainting” , a new function of its text-to-image AI model DALL-E that lets users generate new visuals that?expand the borders of any given picture. This feature came as a request from members of the DALL-E beta program that has amassed more than one million users including thousands of working artists.

• The Shanghai Conservatory of Music, a prestigious Chinese music institution, “enrolled” a virtual singing student who will join the research and development program for AI technology-assisted music creation at the school. The “singer”, Luya, was created by the conservatory and company iFLYTEK to conduct fusion experiments of different music styles from the perspective of songwriting and arrangement.
• Ge Yue, Apple VP and managing director of Apple Greater China, gave a rare speech at the 2022 World Artificial Intelligence Conference in Shanghai outlining the accessibility focus of their machine learning programs. Assistive Touch, Eye-Tracking, and Voice Over are the latest features combining machine learning and built-in sensor data on Apple's devices to address the needs of users with visual, hearing, motor and cognitive impairments.

Open Source Highlights

• Creators of the programming language Go, an increasingly popular language in machine learning systems, added support for vulnerability management . The project, anchored by a Go vulnerability database, includes Go tools that analyze a codebase and surface known vulnerabilities. These tools reduce noise in the results by only surfacing vulnerabilities in functions that a piece of code is actually calling for and developers are encouraged to?contribute to the project ?and provide feedback.
• The regulation of general-purpose AI (GPAI) is currently being debated by the European Union’s legislative bodies as they work on the Artificial Intelligence Act (AIA). Concerns are being raised by a proposal to regulate open-source GPAI which experts warn could further concentrate power over the future of AI in large technology companies and prevent research that is critical to the public’s understanding of AI.
• Meta announces Velox , an open-source execution engine aimed at accelerating data management systems and streamlining their development. This project will attempt to solve the challenge of fragmentation within systems with different feature sets and inconsistent semantics. With Velox, the data-intensive components of data computation engines are unified while still maintaining the extensibility and adaptability for different computation engines.

Lightning AI Highlights

• Mike Ruberry , a PyTorch core technical lead of the past 4+ years, became the first member and Chief Engineer of the PyTorch Team at Lightning. Because PyTorch powers Lightning, building an in-house team will enable Lightning to ensure that it becomes the framework of choice for AI developers and creators.
• Whether you like to prototype models quickly in?Jupyter notebooks,?Kaggle or?Google Colab, Lightning’s got you covered. With the?release of 1.7 , notebook users get to try a?shiny new strategy that provides a multi-GPU experience, similar to running a regular script.

Community Spotlight

Want your work featured? Contact us on Slack or email us at olya @lightning.ai

• Max Pham ’s project is an end-to-end and automatic Vietnamese speech recognition system . It uses PyTorch Lightning to deploy and extend the project into a simple production scenario, and can be easily modified and used with other languages.
• The AI4Sim team, in collaboration with various partners, is developing deep learning solutions to make physical modeling and simulation more accurate and efficient. Their project uses PyTorch Lightning to deal with boilerplate code, early stopping, tensorboard logging, and training parallelization.

Conferences

• NLP 2022 : 11th International Conference on Natural Language Processing. September 17-18, 2022 (Copenhagen, Denmark).
• GTC 2022 : Developer Conference for the Era of AI. September 19-22, 2022 (San Jose, California).
• MLNLP 2022 : 3rd International Conference on Machine Learning Techniques and NLP. September 24-25, 2022 (Toronto, Canada).
• PyTorch Conference : Brings together leading academics, researchers and developers from the Machine Learning community to learn more about software releases on PyTorch. December 2, 2022 (New Orleans, Louisiana).

Don’t Miss the Submission Deadline

• ODSC West 2022 : Open Data Science Conference November 1-3, 2022 (San Francisco, California)
• CCVPR 2022 : 5th International Joint Conference on Computer Vision and Pattern Recognition December 9-11, 2022 (Kuala Lumpur，Malaysia) - Abstracts Due September 9!

Upcoming Community Events

• Meetup : Speed up your Machine Learning Applications with Lightning AI at PyData Miami / Machine Learning Meetup September 22, 2022

Want to learn more from Lightning AI? “Subscribe” to make sure you don’t miss the latest flashes of inspiration, news, tutorials, educational courses, and other AI-driven resources from around the industry. Thanks for reading!