Raeid Saqur

Vector Institute - 661 University Avenue | PT 263 - D.L. Pratt Building, 6 King's College Road · raeidsaqur[at]cs[dot]toronto[dot]edu

**Quick update** I am co-instructing CSC401: Natural Language Computer (cross-listed with grad CSC2511) this semester (i.e. W24).

**Current Research Focus** 1. Adaptive RL agents with world knowledge (pretrained LLMs) playing financial market games using events knowledge from multi-modal sources. 2. Can you enable self-play in such games, or, simulate the financial market with theoretically sound bounds? 3. Areas of interest: RL, More RL, IRL, Reward Motifs, NLP, Multi-modal graph representations, and LLMs (doh).

I am currently a Fulbright Scholar at Princeton University, Department of Computer Science, RBC Fellow and 4th year PhD candidate at the Department of Computer Science (DCS), University of Toronto (UofT) and Vector Institute of AI.

My research interests include natural language understanding (NLU), (multi-modal) semantic representation, probabilistic graphical models (PGMs), artificial general intelligence (AGI). My research philosophy is driven by the imminent need for 'non-expert' human users communicating meaningfully with ubiquitous AI agents using natural language interfaces. Or, from a different lens, long-horizon decision making by AI agents from natual language observations. Most recently, I've been exploring how deep RL agents can dynamically adapt to (hidden) state transitions based on belief states about the environment from natual language observations.

Past: MIT CSAIL; IBM Redbooks (Research Triangle Park, Raleigh), IBM Research, SWG; MBA @ Rotman School of Management, UofT; @Ecobee (Voice-control (Siri) integration), (Co-)Founder @MutuoHealth, @Finatechal;

Research & Publications

I have an eclectic mix of research experience in various ML sub-domains like NLP, Computer Vision and Reinforcement-Learning. My over-arching research interest/goal lie in endowing AI agents with the ability to autonomously acquire skills for executing complex multi-modal tasks. Specifically, how can AI agents learn from 'non-expert' human feedback using natural interfaces like speech, text, gestures etc.

Selected publications and current papers (in progress) here:

Saqur, Raeid, et al. Filtered not Mixed: Stochastic Filtering-Based Online Gating for Mixture of Large Language Models. 2024. preprint
Limmer, Y., Kratsios, A., Yang, X., Saqur, R., & Horvath, B. (2024). Reality Only Happens Once: Single-Path Generalization Bounds for Transformers. 2024. preprint
NIFTY Financial News Headlines Dataset. Raeid Saqur. 2024. preprint
Learning then Leveraging Structures Help with Complex, Compositional, Causal Sequential Tasks in Inverse Reinforcement Learning. Raeid Saqur. 2024. preprint
Large Language Models are Fixated by Red Herrings: Exploring Creative Problem Solving and Einstellung Effect using the Only Connect Wall Dataset. Naeini Saeid, Saqur Raeid, Saeidi Mozhgan, Giorgi John, Taati Babak. paper, project page
Proceedings of NeuRIPS 2023
Hype, Media Frenzy, and Mass Societal Hysteria: Perspectives on Human-imitative Intelligence. Saqur Raeid, Sarpietro Riccardo, Shaheer Muhammad, Buehler Marcel. paper,
Essay Competition, Best Paper (Hon.), ICVSS 2023, Sicily, Italy
SMIRL: Reward Learning Using Structural Motifs in Inverse Reinforcement Learning. Raeid Saqur, Animesh Garg and Frank Rudzicz. abstract, arxiv
Arxiv Preprint: 2022
Multimodal Graph Networks (MGN): Saqur, Raeid, and Karthik Narasimhan. "Multimodal Graph Networks for Compositional Generalization in Visual Question Answering." Proceedings of NeuRIPS 2020 Advances in Neural Information Processing Systems 33 (2020): 3070-3081. abstract, Code
Ethics of Artificial Intelligence in Surgery abstract
Rudzicz, Frank, and Raeid Saqur. "Ethics of Artificial Intelligence in Surgery." arXiv preprint arXiv:2007.14302 (2020).
CLEVR Parser: A Graph Parser Library for Geometric Learning on Language Grounded Image Scenes. paper, code
Proceedings of Second Workshop for NLP Open Source Software (NLP-OSS). EMNLP (2020)
CapsGAN: Using Dynamic Routing for Generative Adversarial Networks abstract, pdf
In proceedings: Computer Vision Conference (CVC) 2019
ReGAN: RE[LAX|BAR|INFORCE] based Sequence Generation using GANs abstract, pdf
PrivySense: Price Volatility based Sentiments Estimation from Financial News using Machine Learning abstract, pdf
H4-Writer: A Text Entry Method Designed For Gaze Controlled Environment with Low KSPC and Spatial Footprint abstract, pdf

Published as Mobile Enterprise Thought Leader, IBM Redbooks, Research Triangle Park, Raleigh, NC:

Saqur, R. (2014, February 5). Mobile web app: Five practices for handling device orientation changes. IBM Developer Works, The Mobile Frontier. link
Saqur, R. (2014, Jan 20). Four reasons for using Node.js to simulate mobile business services layer. IBM Developer Works, The Mobile Frontier. link
Saqur, R. (2013, November 27). IBM Worklight: Using LESS for smarter styling. IBM Developer Works, The Mobile Frontier. link

Industry Relevant R&D Work::

Saqur, Raeid (2016). Richr, RichrPro iOS Apps. Apple AppStore. (R&D work at Finatechal Inc.) Apple iTunes
Saqur, Raeid (2017) Richr Android App. Google PlayStore. . (R&D work at Finatechal Inc.) Google Playstore
Saqur, Raeid (2016). Open-source library: objective-C iOS Homekit scaffolding for IoT devices. (R&D work at Ecobee Inc.) Github
Saqur, Raeid (2015). HomeKit and Alexa enabled Ecobee3 Smart Thermostat with voice command interface. (R&D work at Ecobee Inc.) Apple iTunes
Saqur, Raeid (2014). Bell OrderMax Guided Process for Order Completion. 65% efficiency increase in 6000 call-centers globally, resulting in $2M decreases bottom line. (R&D work at IBM Canada).
Saqur, Raeid (2013). Pioneering work on mobile Cheque Deposit, Cheque-InTM, on iOS and Android platforms and contributing to open-source Tesseract OCR engine. (R&D work at IBM Canada).
Saqur, Raeid (2012). TD Waterhouse Wealth management SDK and Mobile Web Brokerage. Awarded ‘Best of IBM’ award in recognition of industry impact and innovation. (R&D work at IBM Canada)
Saqur, Raeid (2011). Novel hybrid embedded web carousel for native mobile apps. Aeroplan.
Saqur, Raeid (2009). IBM WBP merger algorithm and visualizer. (Engineering Intern at IBM SWG).