AI is changing the is changing the span and scope of human innovation. From creating generated text, images and video to synthesizing the sequences of new antibodies and proteins, AI is expanding the horizons of what humanity can achieve in a lifetime. Join us to hear about the exciting frontier of AI-assisted innovation, which I call Artificial Generative Innovation.

Technology has changed the way science done from the use of word processing tools for paper writing and simulation modeling for validation to computer-aided proofs in mathematics and the use of web search to speeding up literature reviews. Generative AI is the next in the line of tools and it will be interesting to think about its implications in diverse areas of science. Generative AI is already being used to “polish” papers. The panel will explore other uses of Generative AI and the research problems as well as practical challenges arising out of them.

Amazon delivers tens of billions of packages to customers annually, which requires a large logistics setup. ML based systems could be leveraged to optimize the logistic network for improved customer experience. Some of the problems that could use an ML based solution include delivery date estimation, shipping cost estimation, and proactive identification of logistic defects. The data generated by the logistics system share a common theme of being structured in nature, which could be solved using a common ML framework. This talk will discuss how self-attention models are being used at Amazon to solve structured dataset problems in the logistics domain. It will also deep dive into the architecture choices made to deal with the peculiarities of logistics data at Amazon scale. Further, how our models improve over SOTA baselines is also presented.

At Apple, we believe privacy is a fundamental human right. Privacy-preserving ML is a key research area we focus on. In this talk we will share how we applied Differential Privacy to learn popular photos people take at frequently visited locations in order to improve the selection of cover images for Photo Memories, without personally identifiable data leaving their device. This talk will cover how photos are labeled with the on-device model, and how these label-location pairs get encoded into a one-hot vectors, so that random noise can be added. We will explain the secure aggregation protocol that aggregates noised vectors from thousands of devices with a strong Differential Privacy guarantee.

Graphs are general data structures that can represent information from a variety of domains (social, biomedical, online transactions, and many more). Graph Neural Networks (GNNs) are quickly becoming the de-facto Machine Learning models for learning from Graph data and hereby infer missing information, such as, predicting labels of nodes or imputing missing edges.

The main goal of this tutorial is to communicate new research from the TF-GNN team this year, and help practitioners and researchers implement GNNs in a TensorFlow setting.

Tutorial Website (references & material): https://github.com/tensorflow/gnn/blob/main/examples/tutorials/icml_2023/README.md

With the runaway success of Large Language Models, related applications are being rapidly built and deployed. LLM Application developers are faced with some common issues around learning from feedback, prompt selection, response selection, order of prompts, limited context lengths, slow response times.

Reinforcement learning is a promising technology to address these challenges, particularly with the advent of sample-efficient algorithms for personalization and optimization scenarios. At the forefront of these RL solutions is Vowpal Wabbit (VW), an open-source machine learning toolkit and research platform. VW is known for its speed and scalability and continues to grow through the introduction of cutting-edge algorithms.

In the past year we have been working on integrating Vowpal Wabbit with Large Language Model applications, to address some of the aformentioned challenges.

This workshop aims to explore this integration, we'll discuss the following key areas:

• A brief introduction to Vowpal Wabbit and its use of Contextual Bandits learning.
• Strategies to harness the full potential of Vowpal Wabbit in various application scenarios.
• Learned Orchestration with Vowpal Wabbit. Considering the rapid growth of LLMs, there's a promising future in combining these models with VW.

Join us as we delve into this forward-looking exploration. Whether you're a data scientist or a machine learning enthusiast, this workshop will offer valuable insights into building VW and LLM based applications and how it could impact your machine learning applications.

AI has emerged as a prominent catalyst for growth, offering transformative potential across industries. From the development of Large Language Models (LLM) to AI-generated content (AIGC), AI has significantly enhanced efficiency and accuracy. In this session, we will show Colossal-AI's notable advancement in efficient AI model training and implementation across diverse industries. Our innovative approach incorporates cutting-edge technologies like multi-dimensional parallelism, heterogeneous memory management, and adaptive task scheduling. Through extensive research, we have successfully integrated various AI models, including ChatGPT-like model, image model, bioinformatics model and stable diffusion model, into different industries, delivering tangible AI-driven productivity enhancements. We hope our talk can inspire you with fresh ideas and immerse you in the new era of AI.