Multisensory Integration in Spiking Neural Models : Under Wulfram Gerstner at the Laboratory of Computational Neuroscience at EPFL. Here, I analysed a generative model with a new learning rule that combines principles from variational learning and reinforcement learning, which has the form of a local Spike Timing Dependent Plasticity rule modulated by global factors. I incorporated a novel learning rule on an architecture, modeling multisensory integration. The model was able to learn stationary patterns well but did not give satisfactory results in time for dynamic patterns. We identified the problems as (1) the approximation involved in discretising integration factors in the learning algorithm might lead to errors building up and (2) the time scale required to learn and tune hyperparameters was not adequately judged. A short presentation on it can be found here.
Hallucinations in Garden Path Sentences : Under Amitabha Mukerjee at the Department of Computer Science and Engineering, IIT Kanpur. Here, we've studied sentences that are not traditionally garden path, but are interpreted so because of some priming factors. Behavioural (gaze tracking experiments) ensue to test our hypothesis of linguistic systems employing feedback and active revision and of re-analysis often being lazy and partially correct. A report can be found here.
Adaptation of Biochemical Protocols to Handle Technology Change for Digital Microfluidics : Under Bhargab Bhattacharya at the Advanced Computing and Microelectronics Unit, ISI Kolkata. The objective of the work was to suggest new ways to deal with technology enhancement like processor speed increases in the context of digital microfluidic biochips (DMFB). We developed a method, based on symbolic encoding and SAT solvers, added with rich graph-theoretic procedures, to intelligently use the existing action sequences and save costly resources. The findings were published in a paper available here.
Neurobiology of Olfaction Debate : Here, we debated on whether the vibration theory of olfaction made sense or not. A report can be found here.
Data Analysis to understand control of a robotic arm via neural recordings : As part of a course on Data Analysis and Model Classification, we looked at various ways to predict the movement of a robotic arm after recording from the brain of a monkey who is made to move the robotic arm. A short report can be found here.
Daniel J. Amit : I am currently fascinated by this guy, would be happy to share a scanned copy of a short biography of his if asked.