Generalized epilepsy is a common neurological disorder characterised by sudden abnormal brain activity, called a seizure, leading to uncontrolled physical conditions like staring, jerking, stiffening, or loss of consciousness, resulting in seizure-related accidents. These seizures are managed and treated with drugs. The anti-seizure drugs are developed by preclinical and clinical studies involving analysis of Local Field Potential (LFP) recordings from the brain by identifying biomarkers for the development of epilepsy. The analysis is usually done manually, making it a time-consuming process. Automating the manual process is challenging as the LFP recordings are nuanced and require temporal context for seizure detection and analysis due to the variable morphology of seizures in different subjects. To effectively capture these nuances and model long-term dependencies in LFP signals, the application of Large Language Models (LLMs) is an emerging focus of research and development. Hence, as a contribution to this advancing field of research, we present an explainable Generative AI framework, EpiLiteGPT. We present a pipeline under this framework that detects generalized seizures in a Dravet Syndrome mouse model. The pipeline classifies intracranial EEG (iEEG) segments as normal, artefact, or seizure, and the Seizure Insight Module (SIM) then carries out seizure detection based on these classifications. Furthermore, the pipeline is optimized for hardware resource efficiency, gearing towards potential real-time edge deployment. The pipeline achieves an average seizure event detection sensitivity of 81.5% across 6 subjects (2 Training Mice, 4 Held-Out Mice), while the hardware optimizations reduce the energy consumption by 85%, speed up the pipeline by 2.86x and reduce the LLM memory footprint by 75% from the baseline implementation. Fundamentally, the pipeline reduces time required for seizure detection by 97% as compared to manual analysis, thereby accelerating epilepsy research. Additionally, the proposed framework introduces a novel curriculum learning strategy for training LLMs on iEEG signals and develops a GPT-2-based backbone with potential for the development of personalised seizure detection devices. In doing so, it contributes to the growing body of research on LLMs for seizure detection and their prospective deployment on edge devices. ... Read more

The rapid advancement of neural network applications, including multilayer perceptrons (MLP) and deep convolutional neural networks (CNN), has revolutionized domains such as image recognition, speech processing, and classification. However, the increasing depth and complexity of neural network workloads impose significant computational and energy demands. Conventional hardware, such as CPUs and GPUs, faces growing challenges in meeting these demands due to the "memory wall" problem and the diminishing benefits of Moore’s law and Dennard scaling. These issues are particularly pronounced in edge devices where power and energy efficiency are critical.

To address these limitations, computing-in-memory (CIM) architectures, particularly memristor-based crossbar arrays, have emerged as promising solutions. CIM reduces data movement by performing computations directly within memory, significantly improving energy efficiency and performance. Memristor crossbars excel in analog matrix-vector multiplication (MVM), a fundamental operation in neural networks, making them an ideal candidate for accelerating neural network workloads.

This thesis proposes an end-to-end compilation framework that automates the translation and optimization of neural networks for CIM architectures. The framework supports converting high-level MLP models represented in PyTorch into low-level instructions optimized for crossbar-based spatial CIM architectures. Comprehensive experiments explore the impact of various quantization schemes and design space parameters, revealing trade-offs between performance, energy efficiency, and resource utilization. The results demonstrate the framework's potential to support diverse neuromorphic systems and facilitate the efficient deployment of neural networks on CIM architectures. ... Read more