- Ai - Engineering Mailing Lists

AI Seminar : June 9, 2023
by Mangannavar, Rajesh Devaraddi 08 Jun '23

08 Jun '23

Dear all, Our next AI seminar on *"Lipstick on a Pig: Using Language Models as Few-Shot Learners" *by Sameer Singh is scheduled to be on June 9th (Tomorrow), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ) *Location: * https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta… Please note that this is a Zoom-only event *Lipstick on a Pig: Using Language Models as Few-Shot Learners* Sameer Singh Associate Professor, Computer Science University of California, Irvine *Abstract:* Today's Natural Language Processing (NLP) strategies are heavily reliant on pre-trained language models due to their ability to deliver semantically rich representations. While these models provide impressive few-shot natural language understanding and reasoning capabilities, simply using them as "fill in the blank" prompts may not be a one-size-fits-all solution. Despite the allure of direct application of these models - an allure that grows with model and dataset sizes - the objectives of language modeling and few-shot learning are not perfectly aligned. Unpacking this disparity is crucial. In this talk, I will describe some of our work in characterizing the differences between language modeling and few-shot learning. I will show how language modeling comes with crucial shortcomings for few-shot adaptation and describe a simple approach to address them. Then, focusing on numerical reasoning, I will show that the reasoning ability of the language models depends strongly on simple statistics of the pretraining corpus, performing much more accurately for more common terms. These results suggest language modeling may not be sufficient to learn robust reasoners and that we need to take the pretraining data into account when interpreting few-shot evaluation results. While language models hold substantial promise, making these 'pigs' presentable with 'lipstick' may require a more comprehensive approach than currently anticipated. *Speaker Bio:* Dr. Sameer Singh is an Associate Professor of Computer Science at the University of California, Irvine (UCI). He is working primarily on the robustness and interpretability of machine learning algorithms and models that reason with text and structure for natural language processing. Sameer was a postdoctoral researcher at the University of Washington and received his Ph.D. from the University of Massachusetts, Amherst. He has received the NSF CAREER award, UCI Distinguished Early Career Faculty award, the Hellman Faculty Fellowship, and was selected as a DARPA Riser. His group has received funding from Allen Institute for AI, Amazon, NSF, DARPA, Adobe Research, Hasso Plattner Institute, NEC, Base 11, and FICO. Sameer has published extensively at machine learning and natural language processing venues and received conference paper awards at KDD 2016, ACL 2018, EMNLP 2019, AKBC 2020, ACL 2020, and NAACL 2022. (https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsameersin…) *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : June 9, 2023
by Mangannavar, Rajesh Devaraddi 06 Jun '23

06 Jun '23

Dear all, Our next AI seminar on *"Lipstick on a Pig: Using Language Models as Few-Shot Learners" *by Sameer Singh is scheduled to be on June 9th (Friday), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ) *Location: * https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Foregonsta… Please note that this is a Zoom-only event *Lipstick on a Pig: Using Language Models as Few-Shot Learners* Sameer Singh Associate Professor, Computer Science University of California, Irvine *Abstract:* Today's Natural Language Processing (NLP) strategies are heavily reliant on pre-trained language models due to their ability to deliver semantically rich representations. While these models provide impressive few-shot natural language understanding and reasoning capabilities, simply using them as "fill in the blank" prompts may not be a one-size-fits-all solution. Despite the allure of direct application of these models - an allure that grows with model and dataset sizes - the objectives of language modeling and few-shot learning are not perfectly aligned. Unpacking this disparity is crucial. In this talk, I will describe some of our work in characterizing the differences between language modeling and few-shot learning. I will show how language modeling comes with crucial shortcomings for few-shot adaptation and describe a simple approach to address them. Then, focusing on numerical reasoning, I will show that the reasoning ability of the language models depends strongly on simple statistics of the pretraining corpus, performing much more accurately for more common terms. These results suggest language modeling may not be sufficient to learn robust reasoners and that we need to take the pretraining data into account when interpreting few-shot evaluation results. While language models hold substantial promise, making these 'pigs' presentable with 'lipstick' may require a more comprehensive approach than currently anticipated. *Speaker Bio:* Dr. Sameer Singh is an Associate Professor of Computer Science at the University of California, Irvine (UCI). He is working primarily on the robustness and interpretability of machine learning algorithms and models that reason with text and structure for natural language processing. Sameer was a postdoctoral researcher at the University of Washington and received his Ph.D. from the University of Massachusetts, Amherst. He has received the NSF CAREER award, UCI Distinguished Early Career Faculty award, the Hellman Faculty Fellowship, and was selected as a DARPA Riser. His group has received funding from Allen Institute for AI, Amazon, NSF, DARPA, Adobe Research, Hasso Plattner Institute, NEC, Base 11, and FICO. Sameer has published extensively at machine learning and natural language processing venues and received conference paper awards at KDD 2016, ACL 2018, EMNLP 2019, AKBC 2020, ACL 2020, and NAACL 2022. (https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fsameersin…) *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : June 2, 2023
by Mangannavar, Rajesh Devaraddi 01 Jun '23

01 Jun '23

Dear all, Our next AI seminar on *"Convergence Analysis Framework for Fixed-Point Algorithms in Machine Learning and Signal Processing" *by Raviv Raich is scheduled to be on June 2nd (Tomorrow), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Convergence Analysis Framework for Fixed-Point Algorithms in Machine Learning and Signal Processing* Raviv Raich Associate Professor Electrical and Computer Engineering Oregon State University *Abstract:* Many iterative algorithms are designed to solve challenging signal processing and machine learning optimization problems. Among such algorithms are iterative hard thresholding for sparse reconstruction, projected gradient descent for matrix completion, and alternating projection for phase retrieval. Such algorithms and others can be viewed through the fix-point iteration lens. In this talk, we will examine the fixed-point view of iterative algorithms. We will present results on the analysis of projected gradient descent for the well-known constrained least squares problem and show how such analysis can be used to optimize the iterative solution. As a special case of this framework, we will examine iterative solutions to the matrix completion problem. Our approach provides a stepping stone to the optimization of acceleration approaches across multiple algorithms. *Speaker Bio:* Raviv Raich received the B.Sc. and M.Sc. degrees in electrical engineering from Tel-Aviv University, Tel-Aviv, Israel, in 1994 and 1998, respectively, and the Ph.D. degree in electrical engineering from the Georgia Institute of Technology, Atlanta, GA, in 2004. Between 1999 and 2000, he was a Researcher with the Communications Team, Industrial Research, Ltd., Wellington, New Zealand. From 2004 to 2007, he was a Postdoctoral Fellow with the University of Michigan, Ann Arbor, MI. Raich has been an assistant professor (2007-2013) and an associate professor (2013-2023) with the School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR. His research interests include probabilistic modeling and optimization in signal processing and machine learning. From 2011 to 2014, he was an Associate Editor for the IEEE Transactions on Signal Processing. He was a Member during 2011–2016 and the Chair during 2017–2018 of the Machine Learning for Signal Processing Technical Committee (TC) of the IEEE Signal Processing Society. Since 2019, he has been a Member of the Signal Processing Theory and Methods, TC of the IEEE Signal Processing Society. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : June 2, 2023
by Mangannavar, Rajesh Devaraddi 30 May '23

30 May '23

Dear all, Our next AI seminar on *"Convergence Analysis Framework for Fixed-Point Algorithms in Machine Learning and Signal Processing" *by Raviv Raich is scheduled to be on June 2nd (Friday), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Convergence Analysis Framework for Fixed-Point Algorithms in Machine Learning and Signal Processing* Raviv Raich Associate Professor Electrical and Computer Engineering Oregon State University *Abstract:* Many iterative algorithms are designed to solve challenging signal processing and machine learning optimization problems. Among such algorithms are iterative hard thresholding for sparse reconstruction, projected gradient descent for matrix completion, and alternating projection for phase retrieval. Such algorithms and others can be viewed through the fix-point iteration lens. In this talk, we will examine the fixed-point view of iterative algorithms. We will present results on the analysis of projected gradient descent for the well-known constrained least squares problem and show how such analysis can be used to optimize the iterative solution. As a special case of this framework, we will examine iterative solutions to the matrix completion problem. Our approach provides a stepping stone to the optimization of acceleration approaches across multiple algorithms. *Speaker Bio:* Raviv Raich received the B.Sc. and M.Sc. degrees in electrical engineering from Tel-Aviv University, Tel-Aviv, Israel, in 1994 and 1998, respectively, and the Ph.D. degree in electrical engineering from the Georgia Institute of Technology, Atlanta, GA, in 2004. Between 1999 and 2000, he was a Researcher with the Communications Team, Industrial Research, Ltd., Wellington, New Zealand. From 2004 to 2007, he was a Postdoctoral Fellow with the University of Michigan, Ann Arbor, MI. Raich has been an assistant professor (2007-2013) and an associate professor (2013-2023) with the School of Electrical Engineering and Computer Science, Oregon State University, Corvallis, OR. His research interests include probabilistic modeling and optimization in signal processing and machine learning. From 2011 to 2014, he was an Associate Editor for the IEEE Transactions on Signal Processing. He was a Member during 2011–2016 and the Chair during 2017–2018 of the Machine Learning for Signal Processing Technical Committee (TC) of the IEEE Signal Processing Society. Since 2019, he has been a Member of the Signal Processing Theory and Methods, TC of the IEEE Signal Processing Society. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : May 26, 2023
by Mangannavar, Rajesh Devaraddi 25 May '23

25 May '23

Dear all, Our next AI seminar on *"Integrated Planning and Reinforcement Learning for Compositional Domains" *by Harsha Kokel is scheduled to be on May 26th (Tomorrow), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Integrated Planning and Reinforcement Learning for Compositional Domains*Harsha Kokel, Research Scientist IBM Research *Abstract:* Many real-world domains involving sequential decision-making exhibit a compositional nature. Such domains possess an inherent hierarchy that allows decision-makers to abstract away certain details and focus on making high-level decisions using abstractions. In this talk, I will present an overview of some of our recent efforts in leveraging a combination of planning and reinforcement learning for such domains. First, I will present our approach to constructing task-specific abstractions from the influence information in the domain. Second, I will talk about using that in a framework for learning efficient and generalizable agents. Finally, I will discuss the difference between the action spaces of two sequential-decision-making formulations (MDP & PDDL task) and propose an approach to bridge that gap. *Speaker Bio:* Harsha Kokel is a Research Scientist at IBM Research. Her research focuses on efficient knowledge-guided learning in structured, relational domains. She is interested in sequential decision-making problems and exploring the combination of planning and reinforcement learning. She earned her Ph.D. at the University of Texas at Dallas. Her research has been published in the top AI/ML conferences including AAAI, IJCAI, and NeurIPS. She is currently co-organizing workshops on bridging the gap between planning and RL; at ICAPS & IJCAI 2023. She also serves as an assistant electronic publishing editor for JAIR. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : May 26, 2023
by Mangannavar, Rajesh Devaraddi 22 May '23

22 May '23

Dear all, Our next AI seminar on *"Integrated Planning and Reinforcement Learning for Compositional Domains" *by Harsha Kokel is scheduled to be on May 26th (Friday), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Integrated Planning and Reinforcement Learning for Compositional Domains*Harsha Kokel, Research Scientist IBM Research *Abstract:* Many real-world domains involving sequential decision-making exhibit a compositional nature. Such domains possess an inherent hierarchy that allows decision-makers to abstract away certain details and focus on making high-level decisions using abstractions. In this talk, I will present an overview of some of our recent efforts in leveraging a combination of planning and reinforcement learning for such domains. First, I will present our approach to constructing task-specific abstractions from the influence information in the domain. Second, I will talk about using that in a framework for learning efficient and generalizable agents. Finally, I will discuss the difference between the action spaces of two sequential-decision-making formulations (MDP & PDDL task) and propose an approach to bridge that gap. *Speaker Bio:* Harsha Kokel is a Research Scientist at IBM Research. Her research focuses on efficient knowledge-guided learning in structured, relational domains. She is interested in sequential decision-making problems and exploring the combination of planning and reinforcement learning. She earned her Ph.D. at the University of Texas at Dallas. Her research has been published in the top AI/ML conferences including AAAI, IJCAI, and NeurIPS. She is currently co-organizing workshops on bridging the gap between planning and RL; at ICAPS & IJCAI 2023. She also serves as an assistant electronic publishing editor for JAIR. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : May 19, 2023
by Mangannavar, Rajesh Devaraddi 19 May '23

19 May '23

Dear all, Our next AI seminar on *"What's wrong with LLMs and what we should be building instead" *by Thomas G. Dietterich is scheduled to be on May 19th (Today), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *What's wrong with LLMs and what we should be building instead* Thomas G. Dietterich Distinguished Professor Emeritus School of Electrical Engineering and Computer Science Oregon State University *Abstract:* Large Language Models provide a pre-trained foundation for training many interesting AI systems. However, they have many shortcomings. They are expensive to train and to update, their non-linguistic knowledge is poor, they make false and self-contradictory statements, and these statements can be socially and ethically inappropriate. This talk will review these shortcomings and current efforts to address them within the existing LLM framework. It will then argue for a different, more modular architecture that decomposes the functions of existing LLMs and adds several additional components. We believe this alternative can address all of the shortcomings of LLMs. We will speculate about how this modular architecture could be built through a combination of machine learning and engineering. *Speaker Bio:* Thomas G. Dietterich (AB Oberlin College 1977; MS University of Illinois 1979; PhD Stanford University 1984) is Distinguished Professor Emeritus in the School of Electrical Engineering and Computer Science at Oregon State University. Dietterich is one of the pioneers of the field of Machine Learning and has authored more than 225 refereed publications and two books. His current research topics include robust artificial intelligence, robust human-AI systems, and applications in sustainability. Dietterich has devoted many years of service to the research community. He is a former President of the Association for the Advancement of Artificial Intelligence, and the founding President of the International Machine Learning Society. Other major roles include Executive Editor of the journal Machine Learning, co-founder of the Journal for Machine Learning Research, and program chair of AAAI 1990 and NIPS 2000. He currently serves as one of the moderators for the cs.LG category on arXiv. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : May 19, 2023
by Mangannavar, Rajesh Devaraddi 16 May '23

16 May '23

Dear all, Our next AI seminar on *"What's wrong with LLMs and what we should be building instead" *by Thomas G. Dietterich is scheduled to be on May 19th (Friday), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *What's wrong with LLMs and what we should be building instead* Thomas G. Dietterich Distinguished Professor Emeritus School of Electrical Engineering and Computer Science Oregon State University *Abstract:* Large Language Models provide a pre-trained foundation for training many interesting AI systems. However, they have many shortcomings. They are expensive to train and to update, their non-linguistic knowledge is poor, they make false and self-contradictory statements, and these statements can be socially and ethically inappropriate. This talk will review these shortcomings and current efforts to address them within the existing LLM framework. It will then argue for a different, more modular architecture that decomposes the functions of existing LLMs and adds several additional components. We believe this alternative can address all of the shortcomings of LLMs. We will speculate about how this modular architecture could be built through a combination of machine learning and engineering. *Speaker Bio:* Thomas G. Dietterich (AB Oberlin College 1977; MS University of Illinois 1979; PhD Stanford University 1984) is Distinguished Professor Emeritus in the School of Electrical Engineering and Computer Science at Oregon State University. Dietterich is one of the pioneers of the field of Machine Learning and has authored more than 225 refereed publications and two books. His current research topics include robust artificial intelligence, robust human-AI systems, and applications in sustainability. Dietterich has devoted many years of service to the research community. He is a former President of the Association for the Advancement of Artificial Intelligence, and the founding President of the International Machine Learning Society. Other major roles include Executive Editor of the journal Machine Learning, co-founder of the Journal for Machine Learning Research, and program chair of AAAI 1990 and NIPS 2000. He currently serves as one of the moderators for the cs.LG category on arXiv. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

1 0

AI Seminar : May 12, 2023
by Mangannavar, Rajesh Devaraddi 11 May '23

11 May '23

Dear all, Our next AI seminar on *"Designing Interactive AI for Writers" *by Ken Arnold is scheduled to be on May 12th (Tomorrow), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Designing Interactive AI for Writers* Ken Arnold Assistant Professor Calvin University *Abstract:* When AI systems (Gmail, ChatGPT, etc.) suggest words, writers frequently appropriate them as their own. Although this interaction can help writers, it also threatens accuracy, usefulness, and even integrity. I will show empirical results from controlled studies that found that predictive text systems nudge writers to conform both their writing and their opinions to the system’s suggestions. Since I conjecture that these threats are inherent to autocomplete-style predictive interactions, I ask: can large language models help writers without casting doubt on their authorship? I will show prototypes that explore intelligence-augmentation approaches for structuring and revising documents. I hope to spark conversation about what visions and values might shape how we design interactions with emerging AI systems. *Speaker Bio:* Ken Arnold (B.S., Cornell; M.S., MIT; Ph.D., Harvard) is an assistant professor of computer science and data science at Calvin University. His research has shown how predictive text interfaces, like those in smartphone keyboards and email apps, can shape the content of what people communicate. He is currently working on intelligence augmentation to help writers craft words that are fully their own. His current research interests include human-AI interaction in communication, creativity, and education. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged

2 1

AI Seminar : May 12, 2023
by Mangannavar, Rajesh Devaraddi 09 May '23

09 May '23

Dear all, Our next AI seminar on *"Designing Interactive AI for Writers" *by Ken Arnold is scheduled to be on May 12th (Friday), 1-2 PM PST (Add to Google Calendar <https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fcalendar.…> ). It will be followed by a 30-minute Q&A session with the graduate students. *Location: *Crop Science Building 122 *Designing Interactive AI for Writers* Ken Arnold Assistant Professor Calvin University *Abstract:* When AI systems (Gmail, ChatGPT, etc.) suggest words, writers frequently appropriate them as their own. Although this interaction can help writers, it also threatens accuracy, usefulness, and even integrity. I will show empirical results from controlled studies that found that predictive text systems nudge writers to conform both their writing and their opinions to the system’s suggestions. Since I conjecture that these threats are inherent to autocomplete-style predictive interactions, I ask: can large language models help writers without casting doubt on their authorship? I will show prototypes that explore intelligence-augmentation approaches for structuring and revising documents. I hope to spark conversation about what visions and values might shape how we design interactions with emerging AI systems. *Speaker Bio:* Ken Arnold (B.S., Cornell; M.S., MIT; Ph.D., Harvard) is an assistant professor of computer science and data science at Calvin University. His research has shown how predictive text interfaces, like those in smartphone keyboards and email apps, can shape the content of what people communicate. He is currently working on intelligence augmentation to help writers craft words that are fully their own. His current research interests include human-AI interaction in communication, creativity, and education. *Please watch this space for future AI Seminars :* https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fengineeri… Rajesh Mangannavar, Graduate Student Oregon State University ---- AI Seminar Important Reminders: -> For graduate students in the AI program, attendance is strongly encouraged.

1 0