December 2023 - Colo - huji-cs listserver

CS Colloquium (01/01): Yftah Ziser (Edinburgh University)
by Sagie Benaim 01 Jan '24

01 Jan '24

Dear all, Next week, we have the pleasure of having Dr. Yftah Ziser give a talk in the colloquium. The seminar will be held on Monday, January 1st at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title*: Democratising Natural Language Processing: Overcoming Language and Domain Barriers in Low-Resource Environments *Abstract:* Natural language processing (NLP) has been revolutionised in recent years to the point where it is an inseparable part of our daily lives. The transition to transformer-based models allows us to train models on vast amounts of text efficiently, proving that scale plays a crucial role in improving performance. Unfortunately, many people worldwide are marginalised from getting access to high-quality NLP models, as the language they speak and the domains they are interested in count for only a tiny fraction of current state-of-the-art models' training sets. This talk will address the challenges, approaches, and opportunities for democratising NLP across different languages and domains by developing methods to improve NLP in low-resource scenarios. I will start by discussing how we can ease distribution mismatches to improve performance using representation learning. However, as NLP models become increasingly present in our lives, improving other crucial aspects beyond their performance, such as their fairness, factuality, and our ability to understand their underlying mechanisms, is essential. Therefore, I will also discuss using spectral methods to remove information from neural networks to reduce undesired attributes, such as bias, to increase fairness where sensitive data is scarce. Finally, I will explore future directions for making these models accessible to a broader audience by improving the aspects mentioned above in low-resource scenarios. *Bio*: Yftah Ziser is a Postdoctoral Researcher at the School of Informatics at Edinburgh University, hosted by Shay Cohen. He focuses on Deep-Learning methods for dealing with the resource bottleneck, which seriously challenges the worldwide accessibility of NLP technology. His research develops methods to improve low-resource models' performance, fairness, and factuality while developing analysis methods for deepening our understanding of them. He co-organised the Domain Adaptation for NLP Workshop at EACL 2021. Before joining the University of Edinburgh, Yftah worked as a research scientist at Amazon Alexa. Yftah obtained his PhD from the Technion, where Roi Reichart advised him.

1 2

CS Colloquium (25/12): Ariel Kulik (Technion)
by Sagie Benaim 25 Dec '23

25 Dec '23

Dear all, Next week, we have the pleasure of having Dr. Ariel Kulik give a talk in the colloquium. The seminar will be held on Monday, December 25th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title*: Advances in Constrained Resource Allocation and Randomized Branching *Abstract:* I will discuss two of my recent works which demonstrate how probabilistic tools can be used to attain simple yet powerful algorithms: 1. Constrained resource allocation problems can often be modeled as variants of the classic Bin Packing and Knapsack problems. The study of these problems has had a great impact on the development of algorithmic tools, ranging from simple dynamic programming to sophisticated linear programs and rounding techniques. I will present a new and simple algorithmic approach for obtaining efficient approximations for such problems, based on iterative randomized rounding of Configuration LPs. This simple approach yields the state of art asymptotic approximation ratio for Vector Bin Packing, matches the state of art ratios for other variants of the Bin Packing problem, and is useful for Multiple Knapsack and its variants. While the approach is simple, its analysis requires profound understanding of probabilistic tools and deep structural insights about the studied problems. 2. Two prominent approaches for coping with NP-hard optimization problems are polynomial-time approximation and parameterized algorithms. In recent years, increasing attention has been given to the hybrid approach of parameterized- approximation algorithms, leading to both upper and lower bounds. I will show how an intuitive randomized branching technique can be used to obtain fast parameterized- approximation algorithms for some fundamental problems in graph theory. The running times of the resulting algorithms can be derived from recurrence relations in two variables, for which there was no known solution. A main technical contribution of this work is the analysis of such recurrences using tools from information theory. *Bio*: Ariel Kulik is currently a postdoctoral research associate at the Technion, hosted by Roy Schwartz. Formerly, he was a postdoc at CISPA, hosted by Daniel Marx. Ariel completed his PhD studies in Computer Science at the Technion, supervised by Hadas Shachnai. His research focuses on the design and analysis of approximation algorithms for constrained submodular optimization and resource allocation problems. He also works on parameterized-approximation algorithms, with emphasis on tools for the analysis of branching algorithms.

1 1

CS Colloquium (18/12): Amir Feder (Columbia University)
by Sagie Benaim 18 Dec '23

18 Dec '23

Dear all, Next week, we have the pleasure of having Dr. Amir Feder give a talk in the colloquium. The seminar will be held on Monday, December 18th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title*: Causally-driven ML for Text *Abstract:* A fundamental goal of scientific research is to learn about causal relationships, yet this aspect remains underrepresented in machine learning. Recent advances, particularly with large language models (LLMs), have primarily focused on leveraging neural architectures to extract correlations from extensive datasets. However, despite their success, correlational predictive models can be untrustworthy: they rely on shortcuts in the data, leading to errors when applied to out-of-distribution settings; and their representation of text lacks interpretability, rendering them unsuitable for scientific inquiry. In this talk, I will show how a causal perspective can mitigate these shortcomings. I will show how by understanding causal relationships in data we can reduce reliance on shortcuts, and present causally-driven methods that leverage causal structures to perform better out-of-distribution. Then, I will briefly describe how to build probabilistic models of text that reveal interpretable latent variables tailored for estimating causal effects. Finally, I discuss my vision for building science-ready text models. *Bio*: Amir Feder is a postdoctoral fellow at the Columbia University Data Science Institute, working with David Blei. He is currently also a visiting faculty researcher at Google Research. He works in the field of machine learning and causal inference, with a focus on text data. His research develops methods that integrate causality into natural language processing (NLP) to improve the reliability of NLP systems, and to facilitate scientific inquiry with text data. He was a co-organizer of the First Workshop on NLP and Causal Inference (CI+NLP) at EMNLP 2021, the Tutorial on Causality for NLP at EMNLP 2022, and the Workshop on Spurious Correlations, Invariance and Stability at ICML 2023. Before joining Columbia, Amir received his PhD from the Technion, where he was advised by Roi Reichart and worked closely with Uri Shalit. In a previous (academic) life, he was an economics, statistics and history student at Tel Aviv University, the Hebrew University of Jerusalem and Northwestern University.

1 1

CS MATAR Colloquium (6/12): Ella Corren (UC Berkeley)
by Sagie Benaim 06 Dec '23

06 Dec '23

Dear all, Next week, we have the pleasure of having Dr. Ella Corren give a talk in the colloquium. The seminar will be held on Wednesday, December 6th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title* Understanding and Developing Governance Frameworks for the Information Economy *Abstract* The design of regulatory frameworks for the information economy has far-reaching implications for consumers, markets, the evolution of technology, and society as a whole. However, there are many open questions regarding the effects of current and proposed regulations, and whether they work (or will work) as intended. I seek to build a capacity to theorize, critically analyze, and measure the real-world implications of laws governing the information economy and improve the design of new governance regimes. In this talk, I will present a new conceptual framework under which I analyze how consent, a central governance mechanism for the information economy, shifts considerable burdens — to collect and process information, to make informed decisions, and eventually to be liable for adverse results — to individuals and away from firms. I will also present an empirical study on the usage of data control rights by consumers under the California Consumer Privacy Act (CCPA) showing that very few consumers are using their rights, a finding which raises doubts about the ability of the rights to achieve their regulatory goals. *Bio* Ella Corren is a Doctoral Candidate (J.S.D.) at UC Berkeley School of Law, expecting to graduate in May 2024. She is also a graduate of the UC Berkeley LL.M. Thesis Track Program (Dean’s List honors), the Hebrew University Faculty of Law (LL.B., *Atuda* Program in Law), and Tel Aviv University Faculty of Law (LL.M., *magna cum laude*). She practiced law for nine years, and among other roles has advised international corporations as well as tech startups on technology transactions, intellectual property matters, and privacy issues.

1 1

CS MATAR Colloqium (4/12): Noam Kolt (Univ. Toronto)
by Sagie Benaim 04 Dec '23

04 Dec '23

Dear all, Next week we have the pleasure of having Dr. Noam Kolt give a talk in the colloquium. The seminar will be held on Monday, December 4th at 14:00. Location: C220. The title, abstract and bio appear below. Looking forward to seeing you, Sagie and Liat *Title* Legal Institutions in the Age of AI: Opportunities, Risks, and Pathways Forward *Abstract * Legal institutions interact with AI in new and surprising ways. While the technology promises to improve access to justice, the reliance on language models and other predictive tools could systematically alter legal processes and re-shape the development of legal rules and values. Meanwhile, broader risks from AI technology, including concerns relating to misuse and deployment in sensitive settings, pose novel challenges for regulators. Legal institutions will need to develop mechanisms for rapidly adapting to sociotechnical change while maintaining robust and verifiable democratic accountability. Experimentation in designing regulations for AI could, if conducted thoughtfully, foster broader legal innovation and pave the way for new approaches to building regulatory infrastructure. *Bio * Noam Kolt is a final-year doctoral candidate at the University of Toronto Faculty of Law. His research focuses on the role of AI in the legal system, the regulation of AI, and anticipatory governance across diverse regulatory contexts. Noam’s research is informed by a range of perspectives and skills, including experience as a research advisor to Google DeepMind and as a member of OpenAI’s GPT-4 red team. Previously, Noam practiced international law and corporate law, and held fellowships at the University of Toronto’s Centre for Ethics and Schwartz Reisman Institute for Technology and Society. Noam’s legal scholarship has been published in the Berkeley Technology Law Journal, Yale Law & Policy Review, Washington University Law Review, International Journal of Constitutional Law, and Melbourne University Law Review.

1 2

2024

2023

Colo December 2023 ----- 2024 ----- November 2024 October 2024 September 2024 August 2024 July 2024 June 2024 May 2024 April 2024 March 2024 February 2024 January 2024 ----- 2023 ----- December 2023 November 2023

Colo December 2023