Soran Ghaderi soran-ghaderi

Soran Ghaderi

I received my MSc in Artificial Intelligence at the University of Essex, supervised by Professor Luca Citi. Previously, I obtained my bachelor's degree from the University of Kurdistan under the supervision of Dr P. Moradi.

During my postgraduate studies, I worked on reasoning in LLMs for code generation, through which, I developed "Neural Integration of Iterative Reasoning (NIR)". It consisted of a separate deep-think stage with self-reflection and a direct integration of thoughts (hidden states) into the LLM's main generation's hidden states using model surgery!
My ultimate objective is to study the cognitive mechanisms underlying intelligence and develop agents capable of reasoning and interacting with the real world.

Research Interests

Open to research collaborations on the following topics; please reach out via email. ([email protected])

My research focuses on overcoming the challenges faced by current AI models, particularly in reasoning, planning, and generalization in complex environments. Key areas of interest include:

Reasoning and Planning
Handling OOD problems
Representation learning
Embodied Intelligent Agents
Generative modeling, multimodal learning, self-supervised and reinforcement learning
Creating specialized networks for memory, goal-directed planning, spatial reasoning, and error detection and conflict monitoring
Transformers and attention mechanisms
Applications: AI for science, robotics, and LLMs

Currently, I am exploring diffusion models, score-based and flow-based models, differential geometry (particularly Riemannian geometry), metric learning, energy-based models, GFlowNets, and joint embedding predictive architectures operating in the latent spaces among others to design generalizable intelligent agents capable of reasoning (i.e. as optimization problem), deliberate planning (system 2-like), and able to handle uncertainty (in the inference-time)