Trending this week: Actionable explainability in machine learning; How to do multi-task learning intelligently; Value-accelerated persistent reinforcement learning (VaPRL).
Every week we analyze the most discussed topics on Twitter by Data Science & AI influencers.
The following topics, URLs, resources, and tweets have been automatically extracted using a topic modeling technique based on Sentence BERT, which we have enhanced to fit our use case.
This week, Data Science and AI influencers on Twitter have talked about:
- ML Updates
- Reinforcement Learning Updates
- Latest Applications of AI
The following sections provide all the details for each topic.
This week, AI & data science influencers have shared some updates on machine learning.
Marcus Borba has shared the research paper Directive Explanations for Actionable Explainability in Machine Learning Applications. This paper investigates the prospects of using directive explanations to assist people in achieving recourse of machine learning decisions. The goal of the paper was to go beyond explaining predictions (with counterfactual explanations), but also explaining how the individual could act to obtain their desired outcome (if possible).
Mike Tamir has shared an article on How to Do Multi-Task Learning Intelligently. Multi-task learning (MTL) consists of training a unique model to make multiple kinds of predictions, e.g. image classification and semantic segmentation. This contrasts with the traditional approach of training a machine learning model to accomplish each unique task known as single-task learning (STL). This article discusses the motivation for MTL as well as some use cases, it gives the advantages of MTL, the difficulties, and the recent algorithmic advances.
Dr. Ganapathi Pulipaka has shared a link to the Github repository called AutoML-Zero by Google Research. AutoML-Zero aims to automatically discover computer programs that can solve machine learning tasks, starting from empty or random programs and using only basic math operations. The goal is to simultaneously search for all aspects of an ML algorithm — including the model structure and the learning strategy — while employing minimal human bias. This repository provides the open-source code for the paper: “AutoML-Zero: Evolving Machine Learning Algorithms From Scratch”.
Reinforcement Learning Updates
Also, numerous pieces of content on reinforcement learning (RL) have been shared by the influencers.
Sergey Levine has shared the research paper Persistent Reinforcement Learning via Subgoal Curricula. This paper introduces the Value-accelerated Persistent Reinforcement Learning (VaPRL) approach, which generates a curriculum of initial states such that the agent can bootstrap on the success of easier tasks to efficiently learn harder tasks. The agent can learn to reach the initial states proposed by the curriculum, minimizing the reliance on human interventions into the learning. This new approach enables a more autonomous acquisition of complex behaviors for diverse agents.
Nando De Freitas has shared a link to the Github repository of Mava: a research framework for distributed multi-agent reinforcement learning by InstaDeep AI. Mava is a library for building multi-agent reinforcement learning (MARL) systems. Mava provides useful components, abstractions, utilities and tools for MARL, and allows for simple scaling for multi-process system training and execution while providing a high level of flexibility and composability.
KDnuggets have shared a link to the DeepMind blog post research titled “Generally capable agents emerge from open-ended play”. This post summarizes the research paper “Open-Ended Learning Leads to Generally Capable Agents” which explores how we could overcome the current limitation of RL in creating AI agents with more general and adaptive behavior. For example, in the context of games, it results in agents learning one game at a time, not being able to react to completely new conditions, and play a whole universe of games and tasks, including ones never seen before.
Latest Applications Of AI
This week Data Science and AI influencers talked about the latest applications of AI, shared tutorials, and the latest applications from the world of AI.
Nige Willson and Bob E. Hayes shared an interesting article on What is Artificial Intelligence and everything we need to know about AI. The article talks about different types of AI and what these types of AI can do.
Ronald Van Loon shared an article on how Intel is using Machine Learning to make GTA V look incredibly realistic. Putting the game through the processes researchers Stephan R. Richter, Hassan Abu Alhaija, and Vladlen Kolten created produces a surprising result: a visual look that has unmistakable similarities to the kinds of photos you might casually take through the smudged front window of your car.