Welcome to NeurIPS 2022: Exploring Cutting-Edge AI Research
Advancing best-in-class large models, compute-optimal RL agents, and more transparent, ethical, and fair AI systems
The thirty-sixth International Conference on Neural Information Processing Systems (NeurIPS 2022) is set to take place from 28 November – 9 December 2022, combining in-person events in New Orleans, USA, and virtual attendance options.
NeurIPS stands as the premier conference in the realm of artificial intelligence (AI) and machine learning (ML). We are thrilled to announce our partnership with NeurIPS 2022 as Diamond sponsors, contributing to the discussion and dissemination of cutting-edge advancements in AI and ML.
Our diverse team at DeepMind will be showcasing 47 papers, including 35 collaborative projects in virtual panels and poster sessions. Here’s a sneak peek into some of the groundbreaking research we will be presenting:
Best-in-class large models
Large models (LMs) have revolutionized the AI landscape, showcasing remarkable capabilities in areas such as language, text, audio, and image generation. We are proud to introduce our 70 billion parameter language model, Chinchilla, which outperforms many larger models like Gopher. This achievement has reshaped the scaling norms of large models, leading to the development of more efficient and effective models. Chinchilla recently received the prestigious Outstanding Main Track Paper award at NeurIPS 2022.
In addition to Chinchilla, we are unveiling Flamingo, a family of advanced few-shot learning visual language models. Flamingo exhibits superior performance in handling multimodal tasks, setting new benchmarks in the industry.
Optimising reinforcement learning
Reinforcement learning (RL) continues to show promise in developing AI systems that excel in a variety of complex tasks. Our latest innovation enhances the decision-making capabilities of RL agents by providing them with an expanded information retrieval framework.
Another highlight is our revolutionary RL agent, BYOL-Explore, which excels in curiosity-driven exploration in visually intricate environments. This agent achieves exceptional performance levels while maintaining simplicity and robustness.
Algorithmic advances
Algorithms play a pivotal role in modern computing, driving efficiency and innovation across various domains. Our research introduces a groundbreaking method for automatic network configuration, significantly outperforming existing solutions in terms of speed and accuracy.
Furthermore, we shed light on the concept of “algorithmic alignment” and its implications for improving out-of-distribution performance, providing valuable insights for optimizing algorithms at scale.
Pioneering responsibly
DeepMind’s commitment to ethical AI development is central to our mission. We prioritize transparency, fairness, and accountability in all our AI systems.
Our work focuses on explaining and understanding complex AI behaviors through causal modeling, ensuring accountability and transparency. Additionally, we introduce innovative measures like counterfactual harm to guide AI agents in making ethical decisions.
In the realm of healthcare, we address fairness and model performance issues caused by distribution shifts. Our new methodologies aim to diagnose and mitigate these challenges, ensuring the deployment of safe and reliable ML technologies in medical settings.
Experience the full spectrum of our groundbreaking research at NeurIPS 2022 here.