MOJO: Leveraging Unlabeled Data for Generalizable Neural Population Decoding
Researchers have introduced MOJO (Masked autOencoder-based JOint training), a novel framework for training spike-tokenizing models used in neural decoding. MOJO uniquely combines self-supervised learning (SSL) with supervised learning (SL) to effectively utilize unlabeled neural data, improving performance, especially in low-data scenarios.
Multi-Expert Routing for Low-Resource Manchu OCR
Researchers have developed a multi-expert routing system designed to improve Optical Character Recognition (OCR) for historical Manchu documents. This system leverages a lightweight image classifier to route pages to specialized OCR models, addressing the challenge of diverse writing styles and limited labeled data.
PAT: A RAG-Based System for Whole-Document Translation
This research paper introduces PAT (Pragmatic Auto-Translator), a RAG-based system designed to move large language models (LLMs) beyond sentence-level translation. PAT utilizes a comparable corpus of authentic longform texts to inform whole-document translation, aiming to produce draft translations that are contextually appropriate for the target language and culture.
Earthquaker-AI: A Retrieval-Augmented Generation Framework with Rubric-Based Assessment for Primary School Earthquake Education
This paper introduces Earthquaker-AI, a hybrid educational framework that integrates a Retrieval-Augmented Generation (RAG) conversational AI assistant with a robotics project. The system aims to improve earthquake preparedness and safety awareness in primary school students by combining hands-on simulation with cognitive and metacognitive learning.