Madhavun Candadai Email & Phone Number

Dissertation title: Bits from Behaviors: Understanding Function Using Information in Embedded, Embodied, and Dynamical Neural NetworksAdvisor: Dr. Eduardo J. Izquierdo• Built computational models of neural networks (deep, recurrent and dynamical) • Optimized networks to perform a variety of tasks using evolutionary and reinforcement learning algorithms• Analyzed optimized networks using statistical and information theoretic tools to develop an understanding of how they operate• Demonstrated that the same neural network can manifest itself as distinct yet overlapping functional networks during the course of performing different tasks (Candadai et. al., 2017, GECCO)• Demonstrated that neural resources can be reused among multiple tasks down to the level of transient neural dynamics i.e. same neural activity can produce different behaviors (Candadai et. al., 2018, Cog. Sci. Conf.)• Currently working on extended existing methods to identify the source of information encoded in neural networks: externally provided by the environment versus generated internally (Manuscript under preparation) Other research activities:• Initiated a cross-functional collaboration with Dr. Tom Froese on a project that demonstrated the significant impact of social interaction on neural dynamics (Candadai et. al., 2019, Front. Psych.)• Co-advised a summer research student on a data analytics project that won the Wells-Fargo data science contest, and the people’s choice best poster prize at the School of Informatics spring 2019 symposium• Currently co-advising one undergraduate student on a project that analyzes the advantages of hybrid approaches to learning (evolutionary + reinforcement learning)• Continuously participate in project design and planning meetings for other research projects in the lab

• Investigated information theoretic properties of unsupervised learning of disentangled representations using beta-variational auto-encoders• Identified impediments to hierarchical reinforcement learning by systematically studying learning outcomes under different goal space designs (Dwiel, Candadai et. al., 2019, ICLR)• Contributed to the team that participated in the WUR Autonomous Greenhouse Challenge• Demonstrated the ability of DDPG to train flexible robots and characterized its informational requirements (Dwiel, Candadai et. al., 2019, IROS)

Complex Adaptive Systems Laboratory

Master’s thesis: Attractor Network based Salient Word Extraction • Unsupervised detection of salient words from text-corpora using attractor networks and Python NLTK.• Research involved building and comparing various models for keyword-detection, leading to a paper at IJCNN’15

• Courses – Intelligent Systems, Software Engineering, Probability and Random Processes and Graphical Network Programming.• Skills developed/used - Problem solving, Java coding and debugging, Project Management.

Department Of Bio-Informatics

• Constructed structural and functional brain networks from DTI data using FSL, and Diffusion Toolkit• Modeled developmental changes in the functional network to construct age appropriate human brain atlases

Chennai Area, India

• Designed and developed of a Web Portal for a leading Insurance provider.• Technologies used - HTML/JSP for front end, Java for business layer and Web Services over SOAP.

Ph.D.

Master Of Science (M.S.)

Bachelor Of Technology (B.Tech.)