This week in AI: Training Becomes More Accessible, Melody-Composing Models Gain Accuracy, Google Expands Bug Bounties To Open-Source and More!

This week in AI: Training Becomes More Accessible, Melody-Composing Models Gain Accuracy, Google Expands Bug Bounties To Open-Source and More!

From mastering the art of conversation to composing sophisticated melodies, researchers this week are training models to compete with the highly cultured. Training is now easier, faster, and cheaper with Lightning 1.7. Toshiba smashes their AI engineer recruiting targets through a 3-year training program that is bearing fruit and Google expands bug bounties to its open source projects. Let’s dive in! 

Research Highlights

  • Researchers in China have proposed a new approach to producing engaging responses via AI on social media. Many open-domain dialogue models pre-trained with social media comments face difficulties trying to produce engaging responses when interacting with real users. Researchers trained the new model, Diamante, on two kinds of human feedback (including explicit demonstration and implicit preference) in order to produce a more authentic “chit-chat dataset”.

No alt text provided for this image

  • Researchers from Microsoft Research Asia released MeloForm, a system that is able to generate melody with musical form using expert systems and neural networks. Typically, neural network-based music generation systems struggle to emulate the way that humans compose music because of a lack of labelled data on musical form. The researchers claim that MeloForm can generate melodies with precise musical form control with 98% accuracy.
  • Want to extend your reinforcement learning model for a research demo or full ML product? Simply clone this flexible template that can be easily extended to include your preferred algorithm.
  • A study led by a team at Monash University demonstrated that an AI model can potentially predict the best anti-seizure medication for patients with newly diagnosed epilepsy. Currently, choosing anti-seizure drugs for a patient is a process of trial and error. This cohort study sought to develop a prognostic model that can help patients avoid risky side affects with a "modest" 65% accuracy.

ML Engineering Highlights

  • Select US Android users can now take advantage of Google’s LaMDA conversational AI model through its Test Kitchen App that allows people to give feedback on the company’s emerging AI tech. Google claims the LaMDA model can decipher a conversation's intent by examining words in a sentence or paragraph to predict what will come next, generating long, open-ended conversations on potentially any topic.
  • Amazon announced that its time-series machine learning based forecasting service, Amazon Forecast, can now run what-if assessments to determine how different business scenarios can affect demand estimates. Using 6 different accuracy metrics, this tool can simulate hypothetical scenarios through what-if analyses to stress test planning assumptions by capturing possible outcomes.
  • Toshiba surpassed its AI engineer recruitment target since the launch of their training program in 2019, which increased its AI engineer headcount from 750 to 2,100 as of this week. As part of its AI push, the Japanese electronics giant also announced the Toshiba AI Governance statement to promote the development, provision, and use of trustworthy AI.

Open Source Highlights

  • Lightning (formerly known as PyTorch Lightning) can now be integrated to Amazon SageMaker’s distributed data parallel library with only one line of code change. This and other new features in the SageMaker training portfolio aim to streamline the path to the cloud for users of native open-source frameworks like Lightning and PyTorch DDP.
  • Google launched a new Open Source Software Vulnerability Rewards Program (OSS VRP) that will pay researchers up to $31K for information on vulnerabilities in open source software projects – with a focus on critical software such as Go and Angular.
  • Researchers in Moscow, Russia released PyTorch Image Quality (PIQ), a usability-centric library that contains the most popular modern IQA algorithms, “guaranteed” to be correctly implemented according to their original propositions and thoroughly verified.

Lightning AI Highlights

  • The release of Lightning 1.7 includes support for Apple Silicon, native FSDP, collaborative training, and a ton of new features. These updates all focus on making training easier, faster, cheaper, and more accessible. Including the work of over 100 contributors who worked on features, bug-fixes, and documentation for a total of almost 500 commits since 1.6.0, this is one of Lightning’s (formerly known as PyTorch Lightning) biggest releases yet. You can download the new release by doing $ pip install Lightning.

Community Spotlight

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

  • Marc Skov Madsen’s contribution adds support for Panel, a reactive Python framework that allows users to easily create web apps. This PR adds support for PanelFrontend, which simplifies the building of Panel Web Apps within Lightning Apps.
  • Mattia Gatti’s guide on improving model validation using TorchMetrics, with PyTorch or PyTorch Lightning models. Properly handling validation metrics is essential to understanding which model states perform better than others. TorchMetrics offers an organized, efficient way to manage your validation metrics.
  • Martin Manullang’s project uses PyTorch Lightning to classify Covid-19 CT scans, chest x-rays, and ocular disease datasets. This project uses real-world medical datasets to help users understand PyTorch Lightning.

Upcoming Conferences

  • INTERSPEECH 2022: 23rd international conference on the science and technology of spoken language processing.
  • September 18-22, 2022 (Incheon, Korea)
  • 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)

Don’t Miss the Submission Deadline

  • 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 Deep Learning NYC Meetup.
  • September 8, 2022
  • 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!

To view or add a comment, sign in

More articles by Lightning AI

Insights from the community

Others also viewed

Explore topics