[et_pb_section fb_built=”1″ _builder_version=”4.27.4″ _module_preset=”default” background_image=”https:\/\/www.qc.cuny.edu\/academics\/cs\/wp-content\/uploads\/sites\/137\/2023\/11\/Students-scaled.jpeg” min_height=”500px” custom_padding=”10px|||||” global_colors_info=”{}”][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.27.4″ _module_preset=”default” custom_padding=”31px||0px|||” global_colors_info=”{}”][et_pb_row _builder_version=”4.23.1″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.23.1″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” border_width_bottom=”1px” border_color_bottom=”#E71939″ global_colors_info=”{}”]<\/p>\n
[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” text_font=”–et_global_body_font||||||||” text_font_size=”16px” global_colors_info=”{}”]<\/p>\n
This colloquium is intended to bring together Computer Science and Data Science researchers in the tri-state area (especially in NYC) and to foster collaboration. We welcome talks on any topic of interest to the CS community, including theory, algorithms, machine learning, and data science. If you are interested in attending in-person or online, or would like to give a talk, please contact Seminar Organizer, Jun Li at jun.li@qc.cuny.edu<\/a>.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][\/et_pb_section][et_pb_section fb_built=”1″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”1. Online Bayesian Learning and Ensembles” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Feb 9th 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]Monday, 02\/09\/2026, 12:15pm – 1:30pm Abstract: <\/strong>Many real-world applications of machine learning require continuous, adaptive learning strategies over the course of deployment. We discuss a unified framework for online and sequential inference and ensembling of Bayesian models. We give particular focus to Gaussian processes, a family of flexible non-parametric models, and show how to construct general streaming estimators, and further show how they can be adapted to decentralized federated and robust learning. We finally discuss the fragility of the typical online ensembling method, Bayesian model averaging, and introduce a principled alternative from optimization theory, online Bayesian stacking.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”2. Understanding Memory and Reasoning in Language Models via Markov Processes” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Mar 2nd 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]Monday, 03\/02\/2026<\/span>, 12:15pm – 1:30pm Abstract: <\/strong>Despite their remarkable empirical success, language models remain poorly understood from a principled machine learning perspective. To this end, in this talk, we present a unified Markovian perspective on memory and reasoning in modern\u00a0language models. We show that the attention mechanism can be formally interpreted as a context-conditioned Markov process, enabling a principled analysis of learning dynamics. Under this view, model memory corresponds to a Markov transition matrix, and incorporating new knowledge can be understood as expanding the state space. This perspective motivates embedding-level update methods for continual learning that achieve sample-efficient knowledge integration with zero catastrophic forgetting. Furthermore, by formulating multi-step reasoning as a Markov process, we analyze reasoning in small language models and explain why standard supervised fine-tuning and reinforcement learning can fail under sparse rewards. Together, these results demonstrate that Markov processes provide a unifying lens for understanding and improving memory, reasoning, and learning in language models.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”3. Leveraging Natural Human Behavior for Efficient and Intelligent AR and VR Systems” admin_label=”Heading” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Mar 25th 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]Wednesday, 03\/25\/2026, 12:15PM – 1:30PM Abstract<\/b>: Augmented and virtual reality (AR\/VR) systems are emerging as a critical computing platform in modern life, with increasing impact across fields like education, healthcare and industrial applications. Despite their growing importance, AR and VR devices operate under strict constraints on latency, energy consumption, and computational resources, making efficient system design a fundamental challenge. A defining characteristic that distinguishes AR and VR from conventional edge devices is their direct and continuous interface with the human user, where perception, attention, and intention fundamentally shape system behavior. By leveraging natural human behavior such as gaze, head motion, and hand interaction as first class signals, AR and VR systems can adapt their computation to what truly matters to the user, enabling selective processing and more efficient use of limited resources.<\/p>\n In this talk, I will present recent progress from my group on efficient AR and VR computing, spanning a broad range of applications including AI, graphics, and tracking, as well as the corresponding hardware and system designs that enable these solutions to be implemented efficiently. Together, these efforts illustrate how human centered system design can unlock new opportunities for building efficient, intelligent, and responsive AR and VR platforms.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”4. The Polar Express: Optimal matrix sign methods and their application to the Muon algorithm” admin_label=”Heading” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Apr 15th 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]Wednesday, 04\/15\/2026, 12:15PM – 1:30PM Abstract<\/b>: Computing the polar decomposition and the related matrix sign function has been a well-studied problem in numerical analysis for decades. Recently, it has emerged as an important subroutine within the Muon algorithm for training deep neural networks. However, the requirements of this application differ sharply from classical settings: deep learning demands GPU-friendly algorithms that prioritize high throughput over high precision. We introduce Polar Express, a new method for computing the polar decomposition. Like Newton-Schulz and other classical polynomial methods, our approach uses only matrix-matrix multiplications, making it very efficient on GPUs. Inspired by earlier work of Chen & Chow and Nakatsukasa & Freund, Polar Express adapts the update rule at each iteration by solving a minimax optimization problem. We prove that this strategy minimizes error in a worst-case sense, allowing Polar Express to converge as rapidly as possible both in the early iterations and asymptotically. We also address finite-precision issues, making it practical to use in bfloat16. When integrated into the Muon training framework, our method leads to consistent improvements in validation loss when training a GPT-2 model on one billion tokens from the FineWeb dataset, outperforming recent alternatives across a range of learning rates.<\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”5. COOL-MC: Formal Verification Meets Explainable AI for Sequential Decision-Making ” admin_label=”Heading” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Apr 20th 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” hover_enabled=”0″ global_colors_info=”{}” sticky_enabled=”0″]Monday, 04\/20\/2026, 12:15PM – 1:30PM Abstract<\/b>: Data-driven policy learning is an established paradigm for obtaining sequential decision-making policies. It encompasses approaches such as reinforcement learning and behavioral cloning. However, learned policies often lack formal guarantees and are difficult to interpret, which limits their applicability in safety-critical scenarios. The open-source tool COOL-MC addresses these challenges by tightly integrating formal verification with explainable AI methods, where insights from one inform and strengthen the other. This allows COOL-MC to verify that a trained policy is safe while shedding light on its decision-making. The tool has been applied across domains ranging from healthcare and job scheduling to multi-bridge maintenance, and scales naturally to emerging paradigms, including LLM-based policies, quantum AI, and multi-agent systems, offering a unified verification and explain ability tool for researchers and engineers. <\/p>\n [\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”6. Informal Caregivers’ Mental Models of Generative Artificial Intelligence-Based Conversational Agents for Problem-Solving” admin_label=”Heading” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text admin_label=”Apr 27th 2026″ _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]Monday, 04\/27\/2026, 12:15PM – 1:30PM Abstract<\/b>: People increasingly use artificial intelligence (AI), including Generative AI-based Conversational Agents (GCAs), to solve daily problems. However, users face challenges in understanding GCA’s capabilities, learning to interact with it, and evaluating its outputs. Designing around mental models of GCA could help address such challenges. We interviewed sixteen informal caregivers of family and friends and asked them to use GCAs on various problem-solving tasks. We identified mental models regarding GCA as (1) a search engine, (2) a generative AI tool, (3) a personal assistant, and (4) a conversational partner. Notably, some participants shifted and modified their mental models of GCAs even during the interactions. We presented how the differences in mental models could influence participants’ evaluations of GCAs’ performance. Our findings contribute to understanding the applications of GCA for daily problem-solving and reveal the design tensions introduced by different mental models. We discuss the implications for the safe and effective use of GCAs.[\/et_pb_text][\/et_pb_column][\/et_pb_row][et_pb_row disabled_on=”on|on|on” _builder_version=”4.27.4″ _module_preset=”default” disabled=”on” global_colors_info=”{}”][et_pb_column type=”4_4″ _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][et_pb_heading title=”4. Domain-Driven Database Design (D\u2074): Standardizing Horizontally with Fully Qualified Domains (FQD) Across a Heterogeneous Database environment” _builder_version=”4.27.4″ _module_preset=”default” global_colors_info=”{}”][\/et_pb_heading][et_pb_text _builder_version=”4.27.4″ _module_preset=”default” text_font_size=”16px” global_colors_info=”{}”]<\/p>\n Monday, 10\/22\/2025<\/span>, 12:15pm – 1:30pm Abstract: <\/strong>Modern enterprises struggle with siloed data, inconsistent standards, and governance gaps that undermine trust and scalability.\u00a0<\/span>Domain-Driven Database Design (D4<\/sup>)<\/b>\u00a0<\/span>offers a disruptive alternative by embedding business semantics directly into the physical data model (PDM) through\u00a0<\/span>Fully Qualified Domains (FQD)<\/b>\u00a0<\/span>and\u00a0<\/span>Fully Qualified Table Names (FQTN)<\/b>. These constructs create a horizontally standardized foundation where domains are reusable, schemas are consistent, and governance is embedded at the database layer.<\/p>\n This talk will explore how D4<\/sup>\u00a0<\/span>integrates with a\u00a0<\/span>Business Domain Glossary (BDG)<\/b>\u00a0<\/span>and how Knowledge Graphs can be leveraged to capture semantic relationships, automate domain discovery, and generate enforceable FQDs. By applying the\u00a0<\/span>Single Responsibility Principle (SRP)<\/b>\u00a0<\/span>from SOLID Design Principles, FQDs encapsulate clear business meaning, default values, and validation rules, driving systemic clarity and reducing redundancy across a\u00a0<\/span>heterogeneous database environment<\/b>\u00a0<\/span>(SQL Server, PostgreSQL, and beyond).<\/p>\n Bio:<\/strong> Peter Heller is an Adjunct Lecturer in Computer Science at ºì¶¹ÊÓƵ (CUNY), where he teaches courses on SQL Server, Business Intelligence, and Data Modeling. He is the creator of\u00a0D4<\/sup>, <\/b><\/b>a metamodeling framework that embeds governance, semantics, and standardization directly into physical database architectures.<\/span><\/p>\n
\nRoom: Science Building, C205
\nSpeaker: Daniel Waxman<\/a>, Basis AI<\/a><\/p>\n
\nRoom: Science Building, C205
\nSpeaker: Yingcong Li<\/a>, New Jersey Institute of Technology<\/a><\/p>\n
\nRoom: Science Building, C205
\nSpeaker:\u00a0Sai Zhang<\/a>,\u00a0NYU<\/a><\/p>\n
\nRoom: Science Building, C205
\nSpeaker:\u00a0David Persson<\/a>,\u00a0Flatiron & NYU<\/p>\n
\nRoom: Science Building, C205
\nSpeaker:\u00a0Dennis Gro\u00df, Institut f\u00fcr Kommunikations- und Pr\u00fcfungsforschung gGmbH, Heidelberg <\/p>\n
\nRoom: Science Building, C205
\nSpeaker:\u00a0Lu Wang<\/a>,\u00a0Steven Institute of Technology<\/a><\/p>\n
Room: Science Building, C205
Speaker: Peter Heller, <\/a>ºì¶¹ÊÓƵ, CUNY<\/a><\/p>\n
Peter previously served as a\u00a0Computer Specialist Level IV \/ Solutions Architect\u00a0for the City of New York’s Department of Citywide Administrative Services, where he managed the $800M EC3 energy-cost control system, earning a\u00a0NYC Excellence in Technology Project Management Leadership Award. He has presented at\u00a0Data Modeling Zone, published in SQL Server Pro, and is an active member of professional groups, including the NYC Erwin Modeling Group and LinkedIn’s Modern Excel community.<\/span><\/p>\n