Abstract : Humans reason about everyday situations by making commonsense-based inferences, derived both from explicitly stated information and implicit, unstated knowledge. In this thesis, I investigate whether NLP models have different aspects of causal knowledge about events and how to improve their understanding of narratives and plans.
Answering questions about why people perform actions in a narrative can test whether NLP systems contain and can effectively apply causal knowledge about events. I introduce TellMeWhy, a dataset concerning why characters in short narratives perform the actions described. An evaluation of then SOTA finetuned models show that they are far worse than humans. To improve models, it is important to understand what aspects of causal knowledge they need and how to best use external sources to inject this knowledge. In KnowWhy, I analyze different ways of injecting knowledge into models, which is difficult since we do not know apriori what type of knowledge will be needed to answer a question, hence requiring a ranking model to pick the most important inference. Results show that this retrieved knowledge helps models of all sizes, thereby improving their understanding of narratives.
Next, I study whether models can reason about causal aspects of plans. I focus on testing whether they understand the underlying causal dependencies reflected in the temporal order of a plan's steps. I introduce CAT-Bench, and find that SOTA models are underwhelming, and that model answers are not consistent across questions about the same step pairs. In their current state, these models cannot yet reliably be used for complex user-facing tasks. I then measure contemporary models' ability to perform user-facing and user-centric plan customization. I introduce the use of semi-symbolic edits in large language model (LLM) based agents and test several multi-LLM-agent architectures for plan customization. While LLMs still lack the ability to understand complex customization hints, my results suggest that LLM-based architectures may be worth exploring further for other customization applications. Finally, I distill complex reasoning capabilities into small language models (SLMs) using synthetic data that reflects a decomposition-then-editing process for plan customization. I demonstrate that explicitly teaching this latent causal reasoning significantly improves the quality of SLM-generated customizations. Overall, my work has improved how well NLP models understand complex reasoning associated with events in different contexts.

Speaker: Yash Kumar Lal

Location: NCS 220 or Zoom https://stonybrook.zoom.us/j/95849648243?pwd=dgPpZtDpgwQrK9z1SaPpNbBifaorzk.1
Join Zoom Meeting
https://stonybrook.zoom.us/j/98079526509?pwd=Wkt5eURhVDN5VE56TUloS2h5V1Jodz09

Meeting ID: 980 7952 6509
Passcode: 949941



Abstract Over the last decade, artificial neural networks have undergone a revolution, catalyzed by better
supervised learning algorithms. However, in stark contrast to young animals (including humans), training such
networks requires enormous numbers of labeled examples, leading to the belief that animals must rely instead
mainly on unsupervised learning. The reason is that most animal behavior is not the result of clever learning
algorithms--supervised or unsupervised-- but is encoded in the genome. Specifically, animals are born with
highly structured brain connectivity, which enables them to learn very rapidly. Because the wiring diagram is
far too complex to be specified explicitly in the genome, it must be compressed through a genomic
bottleneck. I will describe results showing how the genomic bottleneck algorithm can lead to dramatic
compression of networks and better generalization, particularly for rapid transfer in supervised and
reinforcement learning.

Anthony Zador is professor of neuroscience at CSHL.
Predictable Autonomy for Cyber-Physical Systems by Stanley Bak, Safe Sky Analytics

ABSTRACT: Cyber-physical systems combine complex physics with complex software. Although these systems offer significant potential in fields such as smart grid design, autonomous robotics and medical systems, verification of CPS designs remains challenging. Model-based design permits simulations to be used to explore potential system behaviors, but individual simulations do not provide full coverage of what the system can do. In particular, simulations cannot guarantee the absence of unsafe behaviors, which is unsettling as many CPS are safety-critical systems.

The goal of set-based analysis methods is to explore a system's behaviors using sets of states, rather than individual states. The usual downside of this approach is that set-based analysis methods are limited in scalability, working only for very small models. This talk describes our recent process on improving the scalability of set-based reachability computation for LTI hybrid automaton models, some of which can apply to very large systems (up to one billion continuous state variables!). Lastly, we'll discuss the significant overlap of techniques used for our scalable reachability analysis methods with set-based input/output analysis of neural networks.

BIO: Stanley Bak is a computer scientist investigating the predictable design of autonomous cyber-physical systems. He strives to develop practical formal methods that are both scalable and useful, which demands developing new theory, programming efficient tools and building experimental systems. He received a Bachelor's degree in Computer Science from Rensselaer Polytechnic Institute (RPI) in 2007 (summa cum laude), and a Master's degree in Computer Science from the University of Illinois at Urbana-Champaign (UIUC) in 2009. He completed his PhD from the Department of Computer Science at UIUC in 2013. He received the Founders Award of Excellence for his undergraduate research at RPI in 2004, the Debra and Ira Cohen Graduate Fellowship from UIUC twice, in 2008 and 2009, and was awarded the Science, Mathematics and Research for Transformation (SMART) Scholarship from 2009 to 2013. From 2013 to 2018, Stanley was a Research Computer Scientist at the US Air Force Research Lab (AFRL), both in the Information Directorate in Rome, NY, and in the Aerospace Systems Directorate in Dayton, OH. He currently helps run Safe Sky Analytics, a research consulting company investigating verification and autonomous systems, and performs teaching as an Adjunct Professor at Georgetown University.
The International Neuroethics Society (INS) Speaker Series on AI & Consciousness

Abstract: Colln Allen and I noted in our 2008 book Moral Machines: Teaching Robots Right From Wrong, that consciousness, a theory of mind, sociability, situational awareness and embodiment are all supra-rational (beyond reason) capabilities that contribute to making ethical decision Whether any of these can be fully instantiated in machines remains an open question. Nevertheless, moral decision making in the digital age will require an evolution in and refinement of specific skills for both humans and for AI. I call one of these evolutions in moral decision making capabilities tradeoff ethics and another a silent ethics. Aspects of this social, and not just technological evolution, will require research by neuroscientists.

Speaker Bio: Wendell Wallach has an international reputation as an expert on the ethics and governance of emerging technologies, particularly artificial intelligence and biotechnologies. He is also senior advisor to The Hastings Center and a scholar at the Yale University Interdisciplinary Center for Bioethics where he chaired Technology and Ethics studies for eleven years. Wallach's latest book, a primer on emerging technologies, is entitled, A Dangerous Master: How to keep technology from slipping beyond our control. He co-authored (with Colin Allen) Moral Machines: Teaching Robots Right From Wrong. Wallach has been referred to as, a Godfather of AI Ethics.

Pre-register here (required): https://umaryland.zoom.us/meeting/register/sPpiR_drR4-9JYDhI2NhJg
Abstract: In this talk, we take a step back and argue that many varied and seemingly unrelated attacks on the web are actually symptoms of one deeper problem that has existed since the web's inception. Whether it is attacks due to expired domain names, cloaking done by malicious websites, malvertising, or even our growing distrust of the news can be largely attributed back to the issue of stateless linking. Stateless linking refers to the absence of any integrity guarantees between the time that a link for a remote resource was created, to a future time when this link is resolved by web clients. We draw on 10+ years of research to demonstrate how stateless linking and the resulting lack of content integrity is the true culprit for many of our past, current, and likely future web problems. Successfully tackling this one really challenging problem, has the potential of solving many of our web woes.

Bio: Nick Nikiforakis is affiliated with the National Security Institute. He received his PhD in Computer Science from KU Leuven in Belgium. He received his MSc, in Parallel and Distributed Systems and BSc in Computer Science from the University of Crete, Greece. His research focuses on web security and privacy, software security, and intrusion detection.

Location: NCS 120
TITLE: Towards a Theory of Encode/Decoder Architectures by Andrej Risteski of CMU

ABSTRACT: A common choice of architecture in representation learning (i.e., learning a good embedding of the data) is an encoder/decoder architecture, which tries to map a part of the input into a good latent representation (via an encoder), and predict the remaining part of the input (via a decoder). Two common examples are universal machine translation: where one tries to learn to translate between any pair of a set of languages via a common latent language, given paired up corpora for only a part of the pairs; and contextual encoders -- where one tries to predict a part of the image, given the rest of the image.
 
We will give a framework for analyzing the sample complexity of such architectures -- i.e., how many pairs of languages do we need to have paired up corpora for? How many image prediction tasks do we have to solve to get a good representation?