GcI: Graph-concept Infusion for Large Language Models to solve math problems
Duong Ta and Tim Oates, Department of Computer Science and Electrical Engineering, UMBC
While Large Language Models excel at basic mathematics, they struggle with competition-level problems due to error accumulation across long reasoning chains. We introduce Graph-Concept Infusion (GcI), a prompting framework that combines an LLM with a knowledge graph of $\approx$ 1,300 mathematical concepts and a fine-tuned DistilBERT classifier, operating through three stages: planning sub-goals, retrieving relevant concepts and examples, and executing with targeted information injection. Across SVAMP, CHAMP, and MATH benchmarks, GcI consistently outperforms zero-shot, Chain-of-Thought, Tree-of-Thought, and Graph-of-Thought baselines, with GPT-4-Turbo accuracy improving from 41.8\% to 58.1\% on CHAMP and from 71.5\% to 88.6\% on MATH—a 17.1-point gain. The largest improvements occur in combinatorics and sequence tasks, demonstrating that explicit concept grounding effectively reduces error propagation during complex reasoning.
Algorithm Characterization and Implementation for Large Volume, High Resolution Multichannel Electroencephalography Data in Seizure Detection
Tinoosh Mohsenin, Computer Science and Electrical Engineering
Adam Page, Computer Science and Electrical Engineering
Amey Kulkarni, Computer Science and Electrical Engineering
Tim Oates, Computer Science and Electrical Engineering
Sid Pramod, Computer Science and Electrical Engineering
Ubiquitous bio-sensing for personalized health monitoring is slowly becoming a reality with the increasing availability of small, diverse, robust, high fidelity sensors. This oncoming
flood of data begs the question of how we will extract useful information from it. In this paper we explore the use of a variety of representations and machine learning algorithms applied to the task of seizure detection in large volue of high resolution, multi-channel EEG data. We explore classification accuracy, computational complexity and memory requirements with a view toward understanding which approaches are most suitable for such tasks as the number of people involved and the amount of data they produce grows to be quite large. In particular, we show that layered learning approaches such as Deep Belief Networks excel along these dimensions. We also present the implementation of these algorithms on different hardware approaches including Virtex-7 FPGA, GPUs and 65 nm-CMOS ASIC.
Seizure detection from multiple physiological signals
Tinoosh Mohsenin, Tim Oates, Adam Page, and Sid Pramod, Department of Computer Science and Electrical Engineering, UMBC
Ubiquitous bio-sensing for personalized health monitoring is slowly becoming a reality with the increasing availability of
small, diverse, robust, high fidelity sensors. This oncoming flood of data begs the question of how we will extract useful
information from it. In this project we explore the use of a variety of representations and machine learning algorithms applied to the task of seizure detection in high resolution, multichannel EEG data. We explore classification accuracy, computational complexity and memory requirements with a view toward understanding which approaches are most suitable for such tasks as the number of people involved and the amount of data they produce grows to be quite large. In particular, we show that layered learning approaches such as Deep Belief Networks excel along these dimensions.