Curriculum Vitae
Akshay Gautam • PhD Student in Computational Neuroscience
Education
Ph.D. in Computational Neuroscience and Machine Learning
University of Edinburgh
Edinburgh, Scotland
- •Focus: Computational models of neural circuits and machine learning applications
- •Advisor: Dr. Angus Chadwick
BSc (Hons) in Artificial Intelligence and Computer Science
University of Edinburgh
Edinburgh, Scotland
- •Grade: First Class
- •Dissertation: "Neural Unmixing: Decoding Recurrent Dynamics and Structured Inputs"
Experience
Research Assistant - Computational Neuroscience & ML
Institute of Machine Learning, University of Edinburgh
Edinburgh, Scotland
- •Implemented and tested integration of two neural dynamics modeling frameworks to analyze learning-induced changes in neural connectivity
- •Combined the iLQR-VAE and low tensor rank RNN frameworks in JAX for inferring inputs and connectivity changes in nonlinear dynamical systems from neural data
- •Work involved wrestling with fundamental identifiability challenges in separating intrinsic dynamics from external forcing
Research Assistant - NLP
ICSA, University of Edinburgh
Edinburgh, Scotland
- •Conducted machine learning research on multi-label safety classification with imbalanced datasets, designing pipelines to integrate active learning with human annotator feedback.
- •Implemented and fine-tuned transformer models (T5, LLaMA, BERT) using HuggingFace Transformers with LoRA and 8-bit quantization to optimize performance under resource constraints. Developed and compared two classification approaches: generative multi-label (seq2seq) and discriminative binary classification for each safety category.
- •Explored LLM-assisted annotation pipelines (LLMaAA methodology) for scalable data collection, integrating prompt engineering with active learning strategies to improve annotation efficiency.
Research Intern - NLP
Generative AI Laboratory, University of Edinburgh
Edinburgh, Scotland
- •Developed machine learning pipelines to identify annotation quality issues in safety-critical text classification datasets with limited labeled data.
- •Designed and implemented a data quality assessment system using confidence learning to detect mislabeled annotations, with systematic false positive/false negative analysis. Experimented with multiple text embedding approaches (Cohere, Sentence Transformers) and clustering methods to analyze annotation patterns.
- •Developed evaluation frameworks using label confidence scores, ROC-AUC, and precision/recall metrics. Preprocessed and analyzed large-scale datasets using Python (scikit-learn, pandas, NumPy) for downstream modeling tasks.