Pioneering the Future of AI at ICML 2022 Conference
This weekend, the thirty-ninth International Conference on Machine Learning (ICML 2022) is taking place from July 17-23, 2022 at the Baltimore Convention Center in Maryland, USA. This hybrid event brings together researchers from various fields, including artificial intelligence, data science, machine vision, computational biology, and speech recognition, to showcase their latest work in machine learning.
At the conference, the focus will be on supporting workshops and socials run by organizations like LatinX, Black in AI, Queer in AI, and Women in Machine Learning, while also presenting 30 papers, including 17 external collaborations. Here’s a sneak peek into some of the groundbreaking presentations that will be featured:
Revolutionizing Reinforcement Learning
Improving reinforcement learning (RL) algorithms is crucial for developing advanced AI systems. This involves enhancing performance accuracy and speed, transfer and zero-shot learning, and reducing computational costs.
One of the highlighted presentations will demonstrate a new way to apply generalised policy improvement (GPI) over compositions of policies to boost an agent’s performance. Another presentation proposes a novel grounded and scalable approach to explore efficiently without bonuses. Additionally, a method for augmenting an RL agent with a memory-based retrieval process will be introduced, reducing the agent’s reliance on its model capacity and enabling fast and flexible use of past experiences.
Advancements in Language Models
Language plays a significant role in understanding intelligence, both in AI systems and humans. The presentations will explore unified scaling laws and retrieval to build larger language models more efficiently. A new dataset and benchmark with StreamingQA will be introduced to evaluate how models adapt to and forget new knowledge over time, while also addressing the current struggles of pre-trained language models with creating longer texts due to short-term memory limitations.
Innovative Algorithmic Reasoning
Neural algorithmic reasoning is a rapidly evolving area of research with the potential to adapt known algorithms to real-world problems. The conference will introduce the CLRS benchmark for algorithmic reasoning, which evaluates neural networks’ performance on thirty classical algorithms. Additionally, a general incremental learning algorithm that adapts hindsight experience replay to automated theorem proving, and a framework for constraint-based learned simulation, will be presented, showcasing novel directions for solving complex simulation problems in science and engineering.
Explore the full range of work at ICML 2022 here.