AI Innovation Institute

Date: March 11, 2022
Time: 2:40PM EST

Title: Towards Scalable and Efficient Machine Learning as a Service (MLaaS)

Abstract:
Driven by the explosive growth of big data, the sustained advances of
Machine Learning (ML), and the fast evolving of computer system
techniques, the past few years have witnessed a surging demand for
Machine Learning as a Service (MLaaS). MLaaS is an emerging computing
paradigm that facilitates ML model design, training, inference serving
and provides optimized executions of ML tasks in an automated,
scalable, and efficient manner. In this talk, I will demonstrate how
to integrate ML algorithm research and system research in synergy to
address the pressing challenges in MLaaS. I will first share a story
about how our system experience led to a novel large batching
algorithm design that revolutionizes large-scale training. Then I will
tell another story about how our gradient compression algorithm
research helped us to discover overlooked critical features of modern
ML systems and thereby build a compression-aware distributed ML
system. I will also briefly discuss a promising future of harnessing
serverless computing for MLaaS model inference serving. I will
conclude my talk with a discussion of interdisciplinary research and
future plans.

Bio:
Dr. Feng Yan is an Assistant Professor of Computer Science and
Engineering at University of Nevada, Reno (UNR) and director of the
Intelligent Data and Systems Lab (IDS Lab). Dr. Yan received M.S. and
Ph.D. degrees in Computer Science from the College of William and Mary
and worked at Microsoft Research and HP Labs. Dr. Yan's research
bridges the fields of big data, machine learning, and systems. The
focus of his research is on developing methodologies and building
systems that are automated, high-performing, efficient, robust, and
user-centric. Some of his recent research topics include large-scale
distributed deep learning, machine learning as a service (MLaaS),
federated learning, AutoML, serverless computing, and broad topics in
cloud and HPC. Dr. Yan is also dedicated to interdisciplinary research
and has established fruitful collaborations with domain experts in
areas such as health, physics, geography, material science, mechanical
engineering, civil engineering, and innovated big data and AI-driven
approaches for these domains. Dr. Yan and his team are actively
publishing at the most prestigious venues in computer system area
(such as SOSP, SC, HPDC, USENIX ATC, EuroSys, FAST, VLDB, etc.) and
machine learning area (such as NIPS/NeurIPS, KDD, AAAI, etc.). Dr. Yan
and his students are the recipients of the Best Student Paper Award of
IEEE CLOUD 2018, the Best Paper Award of CLOUD 2019, and the Best
Student Paper Award of ITNG 2021. Dr. Yan is the recipient of the NSF
CAREER Award, the NSF CRII Award, the CSE Best Researcher Award, and
has been nominated for the Regents' Rising Researcher Award. Dr. Yan
serves as Social Media Chair of ACM SIGMETRICS. To learn more
information, please visit Dr. Yan's homepage:
https://www.cse.unr.edu/~fyan.

Abstract: AI agents are increasingly deployed in complex, interactive environments, yet their runtime remains a major bottleneck for training, evaluation, and real-world use. Typical agent behavior unfolds sequentially, where each action requires an API call that can incur substantial latency. For example, a game of chess between two state-of-the-art agents can take hours. We introduce speculative actions, a lossless acceleration framework for general agentic systems. Inspired by speculative execution in microprocessors and speculative decoding in LLM inference, our method uses faster models to predict likely future actions and executes them in parallel, committing only when predictions match. We evaluate speculative actions across gaming, e-commerce, and web search environments, and additionally study a lossy extension in an operating systems setting. Across domains, we achieve up to 55% next-action prediction accuracy, translating into substantial latency reductions. Finally, we present a cost-latency analysis that formalizes the tradeoff between speculative breadth and time savings. This analysis enables principled tuning and selective branch launching, to ensure multi-branch speculation delivers practical speedups without prohibitive cost growth.

Speaker: Kehan Qi

Location: Old Computer Science Building - CS2311

AI Institute Seminar Title: A Geometric Understanding of Deep Learning Abstract: This work introduces an optimal transportation (OT) view of generative adversarial networks (GANs). Natural datasets have intrinsic patterns, which can be summarized as the manifold distribution principle: the distribution of a class of data is close to a low-dimensional manifold. GANs mainly accomplish two tasks: manifold learning and probability distribution transformation. The latter can be carried out using the classical OT method. From the OT perspective, the generator computes the OT map, while the discriminator computes the Wasserstein distance between the generated data distribution and the real data distribution; both can be reduced to a convex geometric optimization process. Furthermore, OT theory discovers the intrinsic collaborative--instead of competitive--relation between the generator and the discriminator, and the fundamental reason for mode collapse. We also propose a novel generative model, which uses an autoencoder (AE) for manifold learning and OT map for probability distribution transformation. This AE-OT model improves the theoretical rigor and transparency, as well as the computational stability and efficiency; in particular, it eliminates the mode collapse. The experimental results validate our hypothesis, and demonstrate the advantages of our proposed model.

Abstract: Human gaze behavior is a fundamental cue for understanding social intent, human-machine interaction, and cognitive processes. This thesis addresses the challenges of gaze target estimation (GTE), also known as gaze following, by developing a holistic understanding of gaze in complex environments.

The first part of this work improves GTE performance by introducing Patch-level Distribution Prediction (PDP). Unlike traditional models that rely on strict pixel-wise regression, PDP models gaze as a distribution over patches, which better accounts for annotation variance and bridges the gap between target location and in/out-of-frame prediction. To address the laborious nature of data labeling, the second part presents GCDR, the first semi-supervised method for gaze following. By prompting large Visual Question Answering (VQA) models to generate initial Grad-CAM heatmaps and refining them with a diffusion model, this method achieves high performance with significantly fewer human annotations. The third part expands the applicability of GTE to multi-camera environments. By introducing the Multi-View Gaze Target (MVGT) dataset, along with two novel frameworks for integrating information between multiple views and predicting the gaze target across views, we explore a new direction that overcomes single-view limitations such as face occlusion and out-of-view targets.

Building on these foundations, the final part of this thesis proposes a new direction toward semantic social gaze understanding using next-generation multimodal Large Language Models (LLMs). Rather than focusing solely on geometric gaze target localization, we aim to enrich gaze prediction with semantic and relational interpretation in complex social scenes. To this end, we will leverage existing gaze following datasets to derive social gaze supervision, including mutual gaze and shared attention, and obtain aligned language descriptions of scene-level gaze behaviors. This proposed work will enable the model to not only locate gaze targets but also predict structured social gaze relations among individuals, meanwhile generating a concise natural-language summary describing the dominant gaze interactions. By integrating spatial gaze estimation, social relation reasoning, and language-based scene understanding within a unified multimodal model, this work takes an important step toward a holistic understanding of human gaze behavior in real-world environments.

Speaker: Qiaomu Miao

Time: May 5, 2022, Thursday, 02:00 PM Eastern Time (US and Canada)
Place: New Computer Science (NCS) Room 220, and Zoom

Zoom link: https://stonybrook.zoom.us/j/95948672934?pwd=d3ZDcUJkK3VweFBDVWhIVDhtaFU2Zz09
Meeting ID:  959 4867 2934
Passcode:  082036

Title:  Generative Adversarial Learning using Optimal Transport

Abstract: 

Generative Adversarial Learning (GAL) aims to learn a target distribution in an adversarial manner. A Generative Adversarial Network (GAN) is a concrete implementation of GAL using a discriminator and a generator that play a min-max game. GANs have been used in many machine learning and computer vision applications. However, GANs are known to be hard to train, mainly because a min-max saddle point optimization problem needs to be solved in GAL. In this thesis, I investigate several methods to improve generative adversarial learning using Optimal Transport (OT). 

Previous Wasserstein GANs (WGANs) do not compute the correct Wasserstein distance to train the discriminator. To address this problem, I propose WGAN-TS that uses the L1 transport cost and computes the correct Wasserstein distance to train the discriminator. To ensure the local convergence of WGANs, I propose WGAN-QC that adopts the quadratic transport cost. I prove that WGAN-QC not only computes the correct Wasserstein distance but also converges to a local equilibrium point. To compute the Wasserstein distance over the whole dataset, I propose to use Semi-Discrete Optimal Transport (SDOT) to match noise points and the real images during GAN training. To measure the quality of an SDOT map, I use the Maximum Relative Error (MRE) and the L_1 distance between the target distribution and the transported distribution obtained by an OT map. I propose statistical methods to estimate the MRE and the L_1 distance. I propose an efficient Epoch Gradient Descent algorithm for SDOT (SDOT-EGD). To deal with the 2D special case of GAL, I propose to use OT to learn 2D distributions. In particular, I adopt OT to match persistent diagrams in training a topology-aware GAN and learn density maps in the crowd counting task. Finally, I use OT and the topological maps of the crowd to improve the crowd counting performance and propose a topology-based metric to measure the quality of the crowd density maps.

Google Docs has been updated to integration Gemini AI. Learn about the integrations that gives you more features within Google Docs. There are also AI Automations that will keep you updated when things happen with documents.

In this session, you will:

1. Learn about the AI features within Google Docs
2. Learn about workflow to be notified of changes to Docs
3. Understand the features that allow AI to work

Register here.

Making sense of Twitter @ Bloomberg presented by Daniel Preotiuc-Pietro

ABSTRACT: The Bloomberg Terminal has provided ways for investors and journalists to sift through and understand the immense volume of tweets and discover financially-relevant content ever since the SEC approved the use of Twitter for company disclosures back in 2013.

In the first part of the talk, I will showcase how tweets impact financial markets and how Bloomberg is using Natural Language Processing methods to identify financially relevant tweets that move the markets. Our processing pipeline feeds directly to clients, journalists in the newsroom and powers several news analytic products offered by the company including trending companies and consumer sentiment for publicly traded equities.

However, understanding user pragmatic intent in individual tweets would allow us to gain deeper insights and enable new applications. I will present several recent research studies focused on understanding intent including identifying complaints and the roles with which vulgarity is used in social media and how these can help improve applications such as sentiment analysis and hate speech detection.

BIO: Daniel Preotiuc-Pietro is a Senior Research Engineer and Team Lead at Bloomberg LP, where he works on analyzing and building models for real-world large scale social media and news mining and information extraction. His research interests are focused on understanding the social and temporal aspects of text, especially from social media, with applications in domains such as Social Psychology, Law, Political Science and Journalism. Several of his research studies were featured in popular press including the Washington Post, BBC, New Scientist, Scientific American or FiveThirtyEight. He is a co-organizer of the Natural Legal Language Processing workshop series. Prior to joining Bloomberg LP, Daniel was a postdoctoral researcher at the University of Pennsylvania with the interdisciplinary World Well Being Project and obtained his PhD in Natural Language Processing and Machine Learning at the University of Sheffield, UK.

Abstract: Theory-internal work on opacity in phonology has been focused on the challenges these interactions present for one theory (rules, constraints) versus another. But there has also been interest in studying the formal, invariant properties of opaque and other process interactions (Chandlee et al. 2018; Bakovic and Blumenfeld 2024), though these works crucially differ in their underlying assumptions. In this talk I will recontextualize Chandlee et al. (2018)'s result that opaque maps are ISL in light of Bakovic and Blumenfeld (2024)'s recent formal typology of process interactions, and this recontextualization will provide an answer to an open question about the k-value of an interaction map. I will then discuss the implications of this collective formal understanding of opacity for a recent model of lexicon and phonological grammar learning (i.e., Hua and Jardine 2021, Chandlee and Jardine to appear).


Speaker: Prof. Jane Chandlee, Associate Professor in the Department of Linguistics at Haverford College

Location: IACS Seminar room.

Last day of finals for 2019 Fall Semester: https://www.stonybrook.edu/commcms/registrar/registration/exams