We’ve all seen significant changes in NLP in the last few years, as models have moved from highly-supervised, recurrent structures into transformers and large-scale, unsupervised models. Now, with the use of LLMs at an all-time high, we have to ask the question – has computer vision stagnated? Join Nate Harada (creator of open-source toolkit Moonshine, perception at Waymo and Cruise) along with moderator Noah Gale (co-founder at Tribe AI) as they discuss the most exciting current developments in computer vision and where the field is headed, including: How we got here – Has computer vision stagnated? The landscape – What’s most exciting in CV right now? What’s coming – Where are the most exciting opportunities in CV? From startup to scale – How to think about investing in labeled data sets, structuring teams, and evaluating success at every stage

Tribe AI CEO Jacyln Rice Nelson interviews Brendan Ittelson, CTO at Zoom, covering Brendan’s career, innovation at scale, trends in digital communications, and – of course – how Zoom is using AI. This conversation details charting a career from technology into leading technical teams, how to build a resilient org in uncertain times, how simplicity enables scale, the role of company mission for a CTO, how Zoom is using machine learning, and what's next in the era of digital communications.

State-of-the-art (SOTA) Generative Models (GMs) can synthesize photo-realistic images that are hard for humans to distinguish from genuine photos. Identifying and understanding manipulated media are crucial to mitigate the social concerns on the potential misuse of GMs. We propose to perform reverse engineering of GMs to infer model hyperparameters from the images generated by these models. We define a novel problem, "model parsing", as estimating GM network architectures and training loss functions by examining their generated images - a task seemingly impossible for human beings. To tackle this problem, we propose a framework with two components: a Fingerprint Estimation Network (FEN), which estimates a GM fingerprint from a generated image by training with four constraints to encourage the fingerprint to have desired properties, and a Parsing Network (PN), which predicts network architecture and loss functions from the estimated fingerprints. To evaluate our approach, we collect a fake image dataset with 100K images generated by 116 different GMs. Extensive experiments show encouraging results in parsing the hyperparameters of the unseen models. Finally, our fingerprint estimation can be leveraged for deepfake detection and image attribution, as we show by reporting SOTA results on both the deepfake detection (Celeb-DF) and image attribution benchmarks.

Understanding the contents of a large corpus of text, like customer reviews or tweets, is a frequently encountered challenge and a significant ML research undertaking. One approach to tackling this problem is topic modeling. We experience it daily - from trending topics on Twitter to key topics in Amazon's product reviews. In this talk, Prem presents a survey of influential methods like LDA and NMF. He then compares them to recent advances in large language models that began with transformers and showcases techniques to apply them, their effectiveness in tackling this problem, and the challenges that come with them. Prem Viswanathan is a member of Tribe AI, a collective of 250+ engineers and data scientists from industry leaders in AI. He’s also the the Co-Founder & CTO of SwiftCX - an intelligent automation platform for customer teams. Prem also serves as an adjunct faculty at Carnegie Mellon University, Pittsburgh. Previously, he was the Director of ML at Artifact.io, a SaaS startup, where he led applied ML research and engineering in NLP. His research focus is on NLP/NLU and Knowledge Graphs. Prem has over ten years of experience in Machine Learning and building ML-infused products at IBM and AWS. More recently, at AWS, he worked alongside Marinus AI, an AI startup that tackles and disrupts human trafficking, to develop a knowledge graph-powered solution to assist and speed up missing person investigations by 4x.

Dave Marra leads AI and Mixed Reality initiatives at Microsoft. As a solutions-focused leader, he's dedicated to leveraging developments in ML technology to revolutionize the productivity of the most technologically underserved segment of the population-the frontline workers. Dave is focused on connecting modern day productivity tools to the people that actually use the tools: soldiers, facility workers, and a range of technicians. In this talk observe why Dave challenges the notion of a paradigm shift away from hardware to privilege software as he describes a deeply interdependent relationship between the two, and increased access and availability to hardware. AI Exchange will also host special guest, Mark Valentine, former US Air Force Commander, and General Manager of National Security at Microsoft, where Mark and Dave were colleagues. Hosted by Scale's Director of Machine Learning and Engineering, Elliot Branson, this is sure to be an unforgettable talk that will open our eyes to the ways AI can elevate the quality of life and productivity for everyday people performing the work that keeps our society running.

Leading organizations across every industry also need to be technology leaders, and today that means leveraging AI. Organizations applying AI don't necessarily have AI expertise, but they are prepared to leverage AI as a strategic asset for their innovation and product strategies. Many of these companies are looking for established use cases to supercharge their products, operations, and business lines. This is where pre-trained models create an opportunity for organizations without the resources and expertise build their own AI. Alexandr Wang, Scale’s CEO and Co-Founder, will outline how Scale’s product investments are helping all types of organizations and teams apply and create their own AI. Wang will also share how the latest advancements in AI allow next-generation models to be more versatile and attuned to specific business needs that weren’t possible before. He will explain why foundation models are the most exciting advancement happening in AI, and how Scale can help advance foundation models, especially as it applies to tuning these models to make them useful for specific tasks and industries including insurance, defense, logistics and many others. Wang founded Scale while a 19-year-old student to help companies build long-term AI strategies with the right data and infrastructure. Under his leadership, Scale has grown to be valued as a $7 billion company serving hundreds of customers across industries from finance to e-commerce to U.S. government agencies.

Experts working on autonomous vehicles will discuss the rapid pace of research and innovation in machine learning, highlighting the most exciting developments over the last few years and how they approach incorporating the constant advances into an autonomous vehicle safely. The panelists are industry leaders working at NVIDIA, Tesla, Waymo, and Wayve. They will also discuss the unique approaches each OEM takes to leverage machine learning in its self-driving stack, with some using end-to-end learning and others preferring modular learning, and each method's advantages and disadvantages. They will also discuss best practices for integrating complicated sensor suites, software, data management, and machine learning with engineering. Other challenges they will cover include collecting large amounts of data, managing, and labeling datasets, integrating ML models with the rest of the self-driving stack, and how to improve the driver continuously. They will also discuss the critical role of simulation in development, and the current state of self-driving cars and the most difficult challenges they face today, including scaling to new environments quickly.

Over the past couple of years, the consumer robotic market has seen rapid adoption due to the pandemic and dropping prices, and the market is expected to grow at a CAGR of almost 31% through 2027. Last year, Amazon announced Astro, a new household robot for home monitoring that combined AI, computer vision, sensor technology, and voice and edge computing. Astro brought new advances to the consumer robotic market around human-robot interaction and multimodal AI. In this fireside chat, Dr. Ken Washington, VP of Software Engineering for Consumer Robotics, will discuss scaling robotics at Amazon and new features in Astro that are novel in the areas of computer vision, perception training, and mapping. Dr. Washington is joined by Vijay Karunamurthy, Head of Engineering at Scale AI, to discuss the future of robotics at Amazon and the new AI technologies evolving with consumer robotics. Before joining Amazon, Dr. Washington was CTO at Ford Motor Company, where his portfolio included propulsion systems, sustainable and advanced materials, additive manufacturing, next-gen vehicle architectures, controls, and automated systems. Dr. Washington also served as Lockheed Martin Corporation’s first chief privacy officer and as CIO for Sandia National Laboratories. In addition to his election to the NAE in 2020, he received the 2012 Black Engineer of the Year Award in Research Leadership.