A discussion on essential elements needed for automating the human thought process and how they are realized in ASP and the s(CASP) goal-directed ASP engine developed in Dr. Gupta's lab. Find out how default rules, expressible in ASP, help solve the explainable AI problem.
Listen to Dr. Shavlik's decades of research on creating machine learning algorithms that accept instruction beyond input-output pairs. Several algorithms and associated experimental results will be presented, along with shortcomings and challenges. The impact and role of the amazing recent large language models, such as OpenAI's GPT-3, which widely "read the web," and hence read concrete facts as well as general knowledge.
This talk will describe the ML Airport Surface Model and discuss how Dr. Coupe and his team developed prediction models that best support FO and ATC decision making with human-in-the-loop. We will show how we leverage a combination of human input and ML models to generate delay predictions which are used to recommend reroute opportunities to the FO. The talk will conclude with lessons learned throughout the deployment cycle.
In this presentation, Dr. Darku will cover a little history of sequential analysis and highlight its applications and importance to today’s business sectors. He will also talk about some recent works in sequential analysis and how businesses can embrace them in their everyday operations.
A brief introduction to the concepts and practices of leveraging knowledge graphs in the industry. Learn different ways organizations can leverage domain expertise and machine learning together for a more knowledge-based approach to AI applications.
Developing an autonomy stack for a highly dynamic environment such as a hospital is not straightforward. In this talk, Dr. Vu will share how we could get from development to deployment of our autonomous UV-Disinfection robots within only six months by leveraging the knowledge of healthcare experts.
In this session, we’ll explore the role Human-centered AI plays in trust and safety of users in the gig economy.
Dr. Jiang will discuss the noisy annotation problem due to expertise errors, i.e., the inconsistencies between different observers due to human subjectivity, using medical image segmentation as an example.
Sungyong Seo and his team at Google Cloud AI propose a novel training method that simultaneously combines human rules and data into deep learning. The relative contributions of rules and data are themselves a learned parameter.
In this talk, Dr. Gupta will present an introduction to Industrial AI, give some real world examples, highlight challenges and lessons learned and point out new research directions and developments.
The commercial fishing industry is extremely harmful to our world's oceans and under-sea ecosystems. In order to reduce the impact of industry on the environment, a more sustainable method of fishing, fixed-net fishing, exists, but hasn't been able to be profitable until now. AI enables fishermen to tell when the most profitable catch is within the net. However, the data needed to training this AI through traditional methods would take years to acquire. Instead, by incorporating generations of fisherman knowledge into the AI system, this system became possible now.
In this talk, Dr. Nguyen will share the key takeaways on how to practice knowledge-driven data-centric AI and talk about the examples while applying this concept.
We introduced the combined system of machine learning models and diagnostic algorithms based on machine expertise to analyze possible causes of detected anomalies. The combined system effectively enhanced the precision of detecting failure events of cold chain equipment by 30% and reduced the daily workload of operation engineers for monitoring equipment condition by 89%.
Panasonic, leader in rechargeable batteries, automotives and avionic systems, industrial systems, is transforming its business through knowledge-driven AI. This is being achieved by leveraging their greatest asset, domain expert knowledge, and incorporating it systematically into the AI systems. This has resulted in higher accuracies at scale across multiple use cases such as Cold Chain Predictive Maintenance, Energy Optimization and Forecasting and overcome some of the practical challenges faced due to data-related issues and data-centric-ML that have plagued AI performance in the past. In this talk, Mr. Kutsumi and Mr. Kawamura will detail out the Panasonic way of building impactful knowledge-driven-AI solutions at scale.
This session is a case study that summarizes the processes, tools, and lessons learned while addressing these challenges over the past five years, building the Spark NLP for Healthcare models in oncology.
TBA
In this talk, Dr. Danks will explore the nature of trustworthiness — what do we want from a trustworthy AI? — and the strengths and weaknesses of different approaches to building AI with respect to producing that trustworthiness.
TBA
Investing early into complexity and intelligence sciences, we present a new model for AI research to investigate the integration of biological and artificial intelligence in the age of information. We designed new institutional structures and an international network of artificial life scientists, driving a new vision of the tech ecosystem through strong fundamentals rooted in academic research.
Take note of the seven questions aligned to the nascent NIST AI Risk Management Framework that we can use to map, measure, and manage the risks and liabilities associated with AI systems.
In general, AI models target to match the human when making a choice, yet when the choices diverge, there is often no secondary method to verify a claim. The question then arises of what the right answer is, and how human and machine can learn from each other in deriving the right answer. This talk will address that question in the context of medical applications.
Calls for trustworthy AI are widespread from governments, industry, academia, and civil society. Purely data-driven approaches to AI development face well-known barriers to trustworthiness, so one might hope that a knowledge-first approach could overcome these challenges. In this talk, Dr. Wing will explore the nature of trustworthiness — what do we want from a trustworthy AI? — and the strengths and weaknesses of different approaches to building AI with respect to producing that trustworthiness. I will argue that all current approaches to building trustworthy AI face significant challenges, but integration of different approaches holds enormous promise.
In this talk Dr. Warden will tell the story of how new research, engineering, and user interface insights are enabling a new world where machines can understand us much better.
This talk will first give an overview of battery health prognostics and then discuss the long-term testing and methodology development efforts led by a team of researchers at Iowa State University and the University of Connecticut.
In this presentation, we discuss opportunities on how ML can impact the field of industrial and environmental reacting flows, present successful applications, and discuss challenges and requirements that are unique within reacting-flow applications
As an increasing amount of internal and external business data become available, it is no longer sufficient to maintain an “analytic” competitive advantage by traditional data analysis. Meaningful utilization of modern AI/ML techniques, coupled with business-focused implementation, is the key recipe to build competitive advantage through data. In this session, Mr. Deng will share an ML implementation example with emphasis on the internal business user experience.
The manufacturing and IoT industries are undergoing a rapid digital transformation by adopting state-of-the-art data infrastructures. However, many companies struggle to realize value from their heavy investments and find their people and processes are completely isolated from the technologies. A new approach to data governance that combines data with domain knowledge is needed in order for these companies to fully achieve the promise of data and machine learning. In this talk, Mr. Gunnam will cover a real-world use-case of a manufacturing firm's knowledge-first data governance using Metaphor—a modern metadata platform created by veterans from LinkedIn, Facebook, Google, and Amazon.
There is a growing concern that humans are being replaced by automation, artificial intelligence, and other technology breakthroughs. Technology innovation is being viewed often as an onslaught on the current workforce. Will AI and automation replace our farmers? AI and humans can co-exist in a symbiotic way. Learn how AI collaboration with farmers not only yields positive business results but also elevates growers' well-being.
In the world of AI, there are two schools of thought: Data-centric AI, and Knowledge/Human-centric AI. With the latter, in addition to data, human intelligence, expertise, scientific knowledge and real-life learning need to be explicitly embedded in AI algorithms, for the models to become more accurate and derived insights to add further value to the human intellect. The first approach is widely followed and easily understood. With the advent of decision science, the second approach is gaining gradual traction and its real potential has just started to be realized by the AI world.