Murthy V. Devarakonda  Murthy V. Devarakonda photo       

contact information

Distinguished Research Staff Member and Manager, IEEE Fellow
IBM T. J. Watson Research Center, Yorktown Heights, NY USA
  +1dash914dash945dash2432

links


profile


Current Research

I am the Principal Investigator of Watson Medical Analytics project developing algorithms, systems, and analysis of patient record data, especially the textual data. Since winning the Jeopardy! tournament, we adapted Watson to the real world challenges. Watson not only demonstrates a remarkable accuracy in deep question-answering but also represents a major breakthrough in software architecture for intelligent systems.

In this medical analytics project of Watson, our goal is to address the cognitive needs of a physician treating a patient. The important scientific and technical challenges we are working on are: patient record summarization based on generated problems list, disease chronology and plans, and enabling extraction of clinical factors from a patient record. We work with major hospitals in understanding the physician's information needs and making sure that this new technology adds value in treatment processes.

Career Summary

In the IBM parlance, I am a Distinguished Research Staff Member & Manager at IBM Research. My long term work is best characterized as transforming results from measurement, monitoring, and analysis into foundational observations, and applying them to systems design and management. Before Watson, I applied this methodology in the areas of IT Discovery and Dependency Analysis, Cloud Computing, Distributed Systems, File Systems, and Storage. Now, I am fascinated by the challenges of unstructured data analytics and the potential to apply Watson to health care.

 

Work that I am proud of

  • Electronic Medical Records application of Watson, where we are creating breakthrough methods to address physicians' cognitive needs and thus improving the patient outcomes
  • Galapagos discovery and dependency analysis, where we used dependency signatures of common software components (e.g. databases) to discover IT dependencies in large scale data center - it is now a GTS Sites & Facilities services offering. See IT Infrastructure and Dependency Discovery Services using Galapagos/ALDM.
  • Calypso File System where we demonstrated file server recovery without explicit state replication and lightweight distributed locking protocols using passive servers
  • Frequency Based Replacement algorithm (for file buffers) with John Robinson, which is the only algorithm that consistently outperforms LRU because of a clever integration of temporal locality and recurrence
  • My thesis work, where I showed measurement-based state-transition models can successfully predict resource usage of a process

 

Education

I received my Ph.D. from University of Illinois at Urbana-Champaign in Computer Science. My undergraduate degree is from Osmania University, Hyderabad, India in Electronics & Communications Engineering.