Guided Generation Group Reading Sessions (formerly Crisp Deep Learning Paper Threading Group)

Introduction

Welcome to the Crisp Deep Learning Paper Threading Group! Our aim is to foster knowledge exchange, collaboration opportunities, and improve our spoken English skills. Here are the rules to ensure a smooth and productive experience for everyone:

Participation

• Commitment: Members are expected to commit to attending weekly sessions and participating actively in discussions.
• Prerequisite Knowledge: Participants should be familiar with basic deep learning techniques such as Gradient Descent, Transformers, and CNNs.

Meeting Structure

• Schedule: Meetings will be held once a week at a time agreed upon by all members.
• Presentation: Each week, one member will present a Deep Learning paper of their choice. The presentation should last between 20-40 minutes.
• Q&A Session: Following the presentation, there will be a 20-30 minute Q&A session where all members can ask questions and discuss the paper.

Paper Selection

• Relevance: The selected paper must be related to Deep Learning.
• Familiarity: Presenters should choose a paper they are familiar with to ensure a thorough and insightful presentation.
• Variety: Members are encouraged to select papers from different subfields of Deep Learning to broaden the group's exposure.

Presentation Guidelines

• Content: Presenters should cover the paper’s motivation, key contributions, methodology, experiments, and results. Presenters should choose a paper they are familiar with to ensure a thorough and insightful presentation.
• Slides: At most one figure per slide.

Hosts:

• Max Ku (https://kuwingfung.github.io/)
• KellyC
• Ray Tam
• Chiao
• Pingbang Hu

Cycle 5

Date: 22/08/2025 - On-going (Every Friday 10pm ET / 7pm PT / Saturday 10am HKT

Topic	Presenter	Slides	Video
W1 (22/8): Research Brainstorm Meeting	Max Ku (UWaterloo)	N/A	N/A
W2 (29/8): Large Language Diffusion Models	George Leung (Novo Nordisk)	N/A	N/A

Cycle 4

Date: 23/05/2025 - 08/08/2025 (Every Friday 8pm ET)

Topic	Presenter	Slides	Video
W1 (30/5): ICLight: Scaling In-the-Wild Training for Diffusion-based Illumination Harmonization and Editing by Imposing Consistent Light Transport	Ethan (Academia Sinica)	N/A	N/A
W2 (7/6): Research Brainstorm Meeting	Ray (Appier)	N/A	N/A
W3 (13/6): Wayformer: Motion Forecasting via Simple & Efficient Attention Networks	Chris (Cardiff)	N/A	N/A
W4 (20/6): Diffusion Policy: Visuomotor Policy Learning via Action Diffusion	Paul Lee (UBC)	Google Drive	N/A
W5 (27/6): Physics of Language Models	Kai-Ling (Duolingo)	N/A	N/A
W6 (4/7): Embodied AI Agents: Modeling the World	Max Ku (UWaterloo)	N/A	N/A
W7 (11/7): Misinformed by Visualization: What Do We Learn From Misinformative Visualizations?	Sam (Penn State)	N/A	N/A
W8 (18/7): OneRec: Unifying Retrieve and Rank with Generative Recommender and Preference Alignment	Ken Chan (TikTok)	N/A	N/A
W9 (25/7): Scalable Influence Function via Sparse Gradient Projection	Pingbang (UIUC)	N/A	N/A
W10 (1/8): Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification	Hsun-Yu (EPFL)	N/A	N/A
W11 (8/8): Dynamic Parametric Retrieval Augmented Generation for Test-time Knowledge Enhancement	CC (FutureNest)	N/A	N/A

Cycle 3

Date: 28/03/2025 - 02/05/2025 (Every Friday 10pm ET)

Topic	Presenter	Slides	Video
W1 (28/3): TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters	Koios (NTHU)	N/A	N/A
W2 (4/4): TheoremExplainAgent: Towards Multimodal Explanations for LLM Theorem Understanding	Thomas (VoteeAI)	N/A	N/A
W3 (11/4): Generative Classifiers Avoid Shortcut Solutions	Andrew (Purdue)	N/A	N/A
W4 (18/4): Research Brainstorm Meeting: PhotoBench	Ethan (Academia Sinica)	N/A	N/A
W5 (25/4): Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models	ChengYi (Academia Sinica)	N/A	N/A
W6 (2/5): Packing Input Frame Context in Next-Frame Prediction Models for Video Generation	Max Ku (UWaterloo)	N/A	YouTube

Cycle 2

Date: 20/12/2024 ~ 14/03/2025 (Every Friday 9pm EST)

Topic	Presenter	Slides	Video
W1 (20/12): Diffusion-Reward Adversarial Imitation Learning	Shih-Yu Lai	N/A	N/A
W2 (3/1): Camera Settings as Tokens: Modeling Photography on Latent Diffusion Models	Ethan (Academia Sinica)	N/A	N/A
W3 (10/1): Chain-of-Thought Prompting Elicits Reasoning in Large Language Models	Jonathan (UWaterloo)	N/A	N/A
W4 (17/1): DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Life	KellyC (UWash)	N/A	N/A
W5 (24/1): [NEW!] Research Brainstorm Meeting	-	N/A	N/A
(31/1) Lunar New Year - No Meeting
W6 (7/2): Planning-Oriented Autonomous Driving	Cyrus (HKU)	N/A	N/A
W7 (14/2): -Canceled-
W8 (21/2): DeepSeek-V3 Technical Report	Chiao (Tesla)	N/A	N/A
W9 (28/2): DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning	Ray (Appier)	N/A	N/A
W10 (7/3): Mastering Board Games by External and Internal Planning with Language Models	Walker (Qualcomm)	N/A	N/A
W11 (14/3): Transformer-Squared: Self-adaptive LLMs	Ivan Lam (CUHK)	Google Drive	N/A

Cycle 1

Date: 20/09/2024 ~ 13/12/2024 (Every Friday 9pm EST)

Topic	Presenter	Slides	Video
W1: Transparent Image Layer Diffusion using Latent Transparency	Ethan (Academia Sinica)	N/A	N/A
W2: Tropical Geometry of Deep Neural Networks	Edisy (Southampton)	N/A	N/A
W3: Towards Robust Control in Visual Generation and Manipulation	Max Ku (UWaterloo)	N/A	N/A
W4: Denoising Diffusion Restoration Models	Aaron (NCU)	N/A	N/A
W5: MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts	Johnny (CUHK)	Google Drive	YouTube
W6: PaliGemma: A Versatile 3B VLM for Transfer	Ray (Appier)	N/A	N/A
W7: FunSearch: Mathematical Discovery from Program Search with LLMs	Jonathan (UWaterloo)	Google Drive	N/A
W8: Training Diffusion Models with Reinforcement Learning	KJ (UIUC)	N/A	N/A
W9: Graph Machine Learning in the Era of Large Language Models (LLMs)	Arthur (Google)	N/A	N/A

Cycle 0 (Trial Session)

Date: 26/07/2024 ~ 13/09/2024

Topic	Presenter	Slides	Video
W1: High -Resolution Image Synthesis with Latent Diffusion Models	Max Ku (UWaterloo)	Google Drive	N/A
W2: Playing Atari with Deep Reinforcement Learning	Jonathan (UWaterloo)	Google Drive	N/A
W3: Deep learning the structural determinants of protein biochemical properties by comparing structural ensembles with DiffNets	Wei-Tse (Oxford)	Google Drive	N/A
W4: BIOCLIP: A Vision Foundation Model for the Tree of Life	Ethan (Academia Sinica)	Google Drive	N/A
W5: Nonnegative Matrix Factorization	Edisy (Southampton)	Google Drive	N/A
W6: Retrieval Augmented Generation for Knowledge Intensive NLP	Cyrus Hei Chan (HKU)	Google Drive	N/A
W7: Improve Mathematical Reasoning in Language Models by Automated Process Supervision	CY (SUTD)	N/A	N/A
W8: Adding Conditional Control to Text-to-Image Diffusion Models	Johnny (CUHK)	Google Drive	YouTube