January 2024 - nlp_reading_group

Reading Discussion Feb 2nd
by Lin, Zihao 31 Jan '24

31 Jan '24

Hi Everyone, In the reading discussion this week, we will discuss: “Localizing and Editing Knowledge in Text-to-Image Generative Models” This paper discussed where the information is stored in a text-to-image diffusion model using causal tracing method. Besides, it propose a editing method called Diff-QuickFix to effectively edit (or delete) a concept in text-to-image models. We may discussion its potential following work such as batch editing, sequential editing or new method for text-to-image or multi-modal model. The reading discussion is at 5-6 pm on Feb 2nd this Friday. You may attend the reading discussion in Room 366 in the Data & Decision Building, or on Zoom (https://virginiatech.zoom.us/j/5961845863) Best Regards, Zihao Link to paper: https://arxiv.org/abs/2310.13730

1 0

Reading Discussion Jan 26th
by Beigi, Mohammad 24 Jan '24

24 Jan '24

Hi Everyone, In the reading discussion this week, we will discuss: “Information Flow Control in Machine Learning through Modular Model Architecture” This paper proposes a secure machine learning framework using a Transformer-based language model with a Mixture-of-Experts (MoE) architecture to control information flow from training data. This approach allows training on sensitive, access-controlled data, with minimal performance overhead and significant improvements in model accuracy. The reading discussion is at 4-5 pm on Jan 26th this Friday. You may attend the reading discussion in Room 366 in the Data & Decision Building, or on Zoom (https://virginiatech.zoom.us/j/5961845863) Best Regards, Mo Link to paper: https://arxiv.org/pdf/2306.03235.pdf

1 0

[nlplab] Reading Discussion on Friday Jan 12
by Ying Shen 09 Jan '24

09 Jan '24

Hi Everyone, In the reading discussion this week, we will discuss: In-Context Learning Unlocked for Diffusion Models <https://arxiv.org/pdf/2305.01115.pdf> This paper investigates an intriguing problem setting involving visual prompting conditioned on both images and text, unlocking in-context learning within diffusion-based generative models. The reading discussion is at *4-5 pm on Jan 12 *this Friday. You may attend the reading discussion in Room 366 in the Data & Decision Building, or on Zoom *(*https://virginiatech.zoom.us/j/5961845863*)*. Enjoy reading! Best wishes, Ying _______________________________________________ nlp_reading_group mailing list nlp_reading_group(a)cs.vt.edu https://mailman.cs.vt.edu/mailman/listinfo/nlp_reading_group _______________________________________________ nlplab mailing list nlplab(a)cs.vt.edu https://mailman.cs.vt.edu/mailman/listinfo/nlplab

1 0

[nlplab] Reading Discussion on Friday Jan 5
by Jingyuan Qi 03 Jan '24

03 Jan '24

Hi Everyone, In the reading discussion this week, we will discuss: Unified Language-Vision Pretraining in LLM with Dynamic Discrete Visual Tokenization <https://openreview.net/forum?id=FlvtjAB0gl> This paper proposed to unify the pre-training objectives for visual and textual tokens while adapting the LLM for vision-language tasks. The reading discussion is at *4-5 pm on Jan 5 *this Friday. You may attend the reading discussion in Room 366 in the Data & Decision Building, or on Zoom <https://virginiatech.zoom.us/j/5961845863>. Enjoy reading! Thank you for your time! Best, ======================================== Jingyuan Qi Ph.D. student, Department of Computer Science Natural Language Processing Lab Virginia Tech (0710) Blacksburg, VA 24061

1 0