This document contains a short summary of my research goals – the areas I wish to delve into during and after my Masters and PhD.
My research is centered around bridging the gap between artificial intelligence and human cognitive abilities by developing AI algorithms and architectures inspired by neuroscience, behavioural science, and psychology. I aim to understand and replicate the intelligence of the human brain, studying its emergence at multiple levels: from individual neurons to complex behaviors. At the neuron level, I am interested in exploring models of spiking neurons and synapses to evaluate their representational capabilities. At the circuit level, I am interested in researching attention mechanisms, memory networks, and recurrent architectures. At the algorithm level, I am interested in investigating biologically plausible learning methods, such as the Forward-Forward algorithm or Predictive Coding. At the architectural level, I am interested in working on hypernetworks while continuing research on adapters and concept-based models. At the behavioral level, I wish to continue exploring memorization in language models and modeling human behaviors, such as opinions and social interactions, while also studying applications of large language models (LLMs) as autonomous agents to explore multi-agent communication and problem-solving.
Currently, I am working towards integrating associative memories into deep learning systems to emulate how humans abstract concepts and generalize to novel situations with the help of concepts memorized in the past. My overarching objective is to develop AI systems that, inspired by human learning, can learn efficiently from fewer samples and adapt rapidly to new scenarios.
My past research focused on multiple aspects of Computer Vision, including Few-Shot Learning for High-Dynamic Range Imaging and GAN Training, Representation Learning, Computational Photography, and Splicing Detection. Transitioning from computer vision, I engaged in NeuroAI research, including concept-based continual learning, human behavior modelling with LLMs, and planning with language models. My core motivation has always been to understand human intelligence, which led me to study insights from neuroscience and psychology. My past research experiences have provided me with a strong foundation in AI fundamentals, methods, and applications, while my academic coursework in neuroscience and neuromorphic AI has allowed me to draw meaningful connections between AI developments and neuroscientific methodologies. Additionally, my exploration of psychology-inspired AI has informed my understanding of the necessary elements that contribute to human-like intelligence.