EagleCam
Improve a video analysis model for identifying the type of food (rodent, lizard, fish) that an eagle is bringing to its nest for the U.S. Fish and Wildlife Services.
Entities in Context
Entities in ContextPartner: Meta Platforms Inc.Participants: Saeed Goodarzi, Nikhil Kagita, Dennis MinnDescription: We revisited the generalization effectiveness of LLMs by focusing on named entities. Named entities are ubiquitous in current Natural Language Understanding benchmarks, yet they have been largely ignored in order to examine the impact on models' reasoning capabilities. We subjected models to the same evaluation data while modifying them to iterate through a large array of named entities from diverse demographics.
Simple Strategies to Select Layers for Fine-Tuning Language Encoders
Simple Strategies to Select Layers for Fine-Tuning Language Encoders Partner: Microsoft, MAIDAP Participants: Gayatri Belapurkar, Saloni Chalkapurkar, Abhilasha Lodha, Yuanming Tao Description: We proposed two-layer selection methods for fine-tuning language encoders that can comprehensively make the transfer learning process for common NLP tasks such as GLUE and SuperGLUE more resource efficient.
Editing Transformer Models with Common Sense Knowledge (EMNLP Conference, Dec. 2023)
Editing Transformer Models with Common Sense Knowledge (EMNLP Conference, Dec. 2023)Partner: Allen Institute for AIParticipants: Anshita Gupta, Debanjan Mondal, Akshay Krishna SheshadriDescription: Memory editing for updating encyclopedic knowledge in transformers has received increasing attention, but it is unclear if these methods can be adapted for nuanced common sense knowledge. In this research, we proposed an adaptation of MEMIT to edit common sense mistakes in GPT-2 Large and XL. We extend editing to various token locations and employ a robust layer selection strategy. Our results suggest a promising path for improving GPT by incorporating context-specific user feedback about common sense through direct model editing as well as fixing and customizing model behaviors using human-in-the-loop-systems.
Generating Metrics for High-Performance Computing Clusters
Partner: Unity DS4CG 2023. Unity is a collaborative, multi-institutional high-performance computing cluster, primarily used for research computing. The Unity project focused on generating useful metrics and analysis for Unity by building a pipeline to a database that could power a live dashboard for Unity’s admin staff. Metrics included unnecessarily idle GPUs, daily and weekly node usage, total resource usage, and wait time. Additionally, a prediction model for wait time at job submission time was built.
Detecting Extreme Speech in YouTube Videos
Partner: Media Cloud The surge in multimodal content shared online, particularly on platforms like YouTube and Instagram, has increased the need for effective extreme and hateful speech detection systems. Current systems often fail to address the nuanced challenges of detecting explicit and implicit hate speech in multimodal contexts, where speech and text combine to convey harmful messages. Media Cloud, an open-source media research platform, helps researchers study news and information flow globally. This DS4CG team worked in collaboration with Media Cloud to focus on advancing multimodal hate speech detection by addressing three key challenges: the lack of comprehensive, human-annotated datasets; the absence of systems capable of analyzing both audio and text data simultaneously; and the need for fine-grained detection of subtle hate speech. The study leverages distinct latent features from audio and text to improve…