About

Dr. Raj Reddy is a University Professor of Computer Science and Robotics and holds the Moza Bint Nasser Chair in the School of Computer Science at Carnegie Mellon University. His academic journey includes a BE degree from Guindy Engineering College of the University of Madras, an MTech from the University of New South Wales, and a Ph.D. in Computer Science from Stanford University. He has been active in artificial intelligence research for over five decades, focusing on areas such as AI, Speech Understanding, Image Understanding, Robotics, Multi-sensor Fusion, and Intelligent Agents. Throughout his career, Dr. Reddy has held significant academic and leadership roles, including serving as the founding Director of the Robotics Institute at Carnegie Mellon from 1979 to 1991 and as the Dean of the School of Computer Science from 1991 to 1999. He has contributed extensively to the development of AI and robotics, and his current research interests include technology in service of society, speech-to-speech translation among Indian languages, voice computing for semi-literate populations, digital democracy, learning sciences, and expanding educational access through digital means. Recognized globally for his contributions, Dr. Reddy is a Fellow of multiple prestigious organizations, a member of the National Academy of Engineering, and has received numerous awards including the ACM Turing Award, the Padma Bhushan, and the Legion of Honor from France. His work emphasizes the societal impact of technology, aiming to create inclusive, accessible, and beneficial innovations.

Research topics

• Artificial Intelligence
• Computer Science
• Speech recognition
• Algorithm
• Mathematics
• Arithmetic

Selected publications

• The promise and perils of artificial intelligence

  TAO. · 2025-12-01

  articleOpen access1st authorCorresponding

  Artificial intelligence (AI) can be defined as the enhancement of human mental capabilities, and the development of AI involves the creation of various intelligence amplifiers that will lead to superhuman capabilities. AI will have many positive outcomes, including eliminating the literacy divide, overcoming language barriers, preventing pandemic shutdowns, and alleviating poverty and hunger. However, AI will also cause job losses, malicious deepfakes and disinformation, killer weapons, and the emergence of other hazards. In general terms, AI will have a profound impact on society and help to enhance the quality of life for everyone on the planet. We need to develop AI tools across every field to improve productivity; We also need to overhaul our educational curriculum so that future generations can use AI tools effectively.

  PublisherDOI

• SNFD: An Adaptive Machine Learning Approach for Network Failure Prediction

  2025-07-04

  article

  Lorem ipsum dolor sit amet, consectetuer adipiscing elit. Ut purus elit, vestibulum ut, placerat ac, adipiscing vitae, felis. Curabitur dictum gravida mauris. Nam arcu libero, nonummy eget, consectetuer id, vulputate a, magna. Donec vehicula augue eu neque. Pellentesque habitant morbi tristique senectus et netus et malesuada fames ac turpis egestas. Mauris ut leo. Cras viverra metus rhoncus sem. Nulla et lectus vestibulum urna fringilla ultrices. Phasellus eu tellus sit amet tortor gravida placerat. Integer sapien est, iaculis in, pretium quis, viverra ac, nunc. Praesent eget sem vel leo ultrices bibendum. Aenean faucibus. Morbi dolor nulla, malesuada eu, pulvinar at, mollis ac, nulla. Curabitur auctor semper nulla. Donec varius orci eget risus. Duis nibh mi, congue eu, accumsan eleifend, sagittis quis, diam. Duis eget orci sit amet orci dignissim rutrum. In today's digital world, network failures can cause significant disruptions, leading to downtime, financial losses, and security vulnerabilities. The Smart Network Failure Detection (SNFD) System is designed to provide a proactive and intelligent solution for monitoring, detecting, and predicting network failures in real time. By leveraging Artificial Intelligence (AI), Machine Learning (ML), and Big Data Analytics, this system continuously analyzes network traffic, detects anomalies, and alerts administrators before critical failures occur. The system collects data from network devices using protocols like SNMP, NetFlow, and sFlow, processes it using ML models such as Random Forest, Isolation Forest, and Long Short-Term Memory (LSTM) networks, and provides real-time insights through interactive dashboards (Grafana, Kibana). When an anomaly is detected, automated alerts are sent via email, SMS, or web notifications, allowing administrators to take immediate action and prevent downtime. The SNFD system is adaptable across various industries, including IT infrastructure, telecom, healthcare, finance, and IoT networks, ensuring seamless and reliable connectivity. By integrating real-time monitoring, predictive analytics, and automation, the proposed system enhances network efficiency, reduces operational costs, and strengthens cybersecurity. This project aims to bridge the gap between traditional network monitoring and AI-driven automation, offering a scalable and Intelligent solution to network failure management.

  PublisherDOI

• Enhanced Deep Fake Image Detection via Feature Fusion of EfficientNet, Xception, and ResNet Models

  2025-01-07 · 2 citations

  article1st authorCorresponding

  In recent years, the rapid advancements in deep learning have led to the creation of realistic fake images, commonly referred to as deep fakes. Detecting these images has emerged as a critical challenge in the fields of cyber-security, digital forensics, and privacy protection. This paper proposes a novel deep fake detection model that utilizes feature fusion from three powerful pre-trained convolutional neural networks (CNNs): EfficientNetB0, Xception, and ResNet50. By leveraging the strengths of each model, discriminative features are extracted to enhance classification accuracy. The approach combines features from multiple networks, which are then processed through dense layers to classify real and fake images. Evaluated on a dataset of real and fake faces, the proposed model demonstrates significant improvements in detection accuracy and generalization compared to traditional single-model approaches. Comprehensive performance metrics, including precision, recall, and F1-score, are reported, highlighting that the ensemble model outperforms conventional CNN models in deep fake detection.

  PublisherDOI

• Enhancing Predictive Modeling of Diamond Prices using Machine Learning and Meta-Ensemble Techniques

  2024-12-19

  article1st authorCorresponding

  Accurate prediction of diamond prices is vital for decision-making in the diamond industry, benefiting all stakeholders. In this study, we employ a comprehensive dataset encompassing various diamond attributes, including cut, color, clarity, carat weight, and engineered features. We explore the predictive capabilities of traditional machine learning (ML) algorithms such as Linear Regression, Ridge Regression, Lasso Regression, Decision Tree Regression, Gradient Boosting Regression, Random Forest Regression, and AdaBoost Regression. Subsequently, we investigate the efficacy of ensemble techniques, including Bagging, Voting, and Stacking, aiming to enhance predictive accuracy and robustness. This research rigorously evaluates and compares these models' performance using established metrics, including Root Mean Squared Error(RMSE), R-squared, and Mean Absolute Error(MAE). Additionally, we suggest a novel strategy that uses ensemble model predictions as features to train a Meta-Model, increasing prediction accuracy even further. This study's results highlight how well machine learning and ensemble methods can predict diamond prices. We clarify the benefits and drawbacks of different modeling techniques through comparison analysis, offering insights into the best techniques for estimating diamond prices. This study gives stakeholders in the diamond business useful information and a data-driven method for predicting prices and making decisions.

  PublisherDOI

• CCT-217: A Landmark Gene Silencing Therapeutic for Weight Loss That Enhances Lean Mass Composition, Reduces Visceral, Subcutaneous Fat and Restores Metabolic Health in Obesity through Browning

  bioRxiv (Cold Spring Harbor Laboratory) · 2024-12-30 · 1 citations

  preprint1st authorCorresponding

  Abstract According to the CDC, between 2017 and March 2020, 41.9% of U.S. adults aged 20 and older had obesity, with 9.2% experiencing severe obesity, affecting over 100 million and 22 million adults, respectively. Obesity prevalence increased from 30.5% in 1999-2000 to 41.9% in 2017-2020, while severe obesity rose from 4.7% to 9.2%. As per CDC projection, by 2030, nearly half of the US adults will be obese, with a quarter of them being severely obese. Current therapeutic approaches predominantly focus on appetite suppression and caloric restriction. This approach results in both fat mass and lean mass loss and thus compromises long term metabolic health. Preserving lean mass during weight loss is crucial. Remodeling of the white fat into beige fat through the process called browning offers a promising alternative to appetite suppression dependent weight loss, and complements with reduction of visceral fat, marrow adipose tissue, improve liver steatosis, enhancing insulin sensitivity, reduced cardiovascular risk and suppression of inflammation. Here, we present CCT-217, a groundbreaking siRNA therapy targeting CB1R and ZFP423 genes, with strong adipose tissue-specific tropism with nil brain and liver penetration in diet-induced obese (DIO) C57BL/6J male mice model (n=7 control, n=5 treated). Mice received 7 subcutaneous doses over 20 days (initial half-dose, followed by once after three days CB1R siRNA 2.0 mg/kg and ZFP423 siRNA 1.8 mg/kg dose). CCT-217 induced browning of the WAT and resulted in a significant 26% reduction in body weight with a modest 12% decrease in feed intake after the 4th dose, preserved normal clinical behavior and nutrition. Lean mass composition significantly improved to 79% versus 62% in controls, while total fat mass decreased by 43.5%. Notably, visceral fat depots, including retroperitoneal (59.4%), mesenteric (54.3%), and gonadal (42.0%) fat showed substantial reductions, alongside a reduction of 61.7% inguinal white adipose tissue (iWAT). Histological analysis revealed extensive browning of iWAT, reduced adipocyte size, and increased adipocyte count, indicating enhanced thermogenesis and adipose tissue remodeling. Liver pathology showed significant improvement, with reduced hepatocellular adipose vacuolation and the absence of fibrosis, inflammation, or macrophage infiltration, highlighting the hepatoprotective effects of CCT-217. Metabolic markers demonstrated a 20% reduction in plasma cholesterol and decreased systemic inflammation (21% reduction in CRP), accompanied by significantly improved leptin sensitivity and favorable trends in insulin sensitivity. Mechanistically, CCT-217 achieved 42-fold silencing of CB1R and 14.5-fold silencing of ZFP423 specifically in iWAT, with no off-target effects observed in liver or brain. In conclusion, CCT-217 emerges as a landmark therapy that effectively reduces visceral fat, enhances lean mass composition, and restores metabolic health through precise, tissue-specific gene silencing. These findings position CCT-217 as a transformative and highly promising next generation therapeutic for addressing obesity and its associated complications.

  PublisherDOI

• Recognition of Two Connected Handwritten Digits Based on User-Defined Algorithm

  International Journal of Innovative Technology and Exploring Engineering · 2020

  1st authorCorresponding
  • Computer Science
  • Artificial Intelligence
  • Computer Science

  The present paper proposes a model for recognizing unconstrained offline two connected handwritten Numeral digit strings. The Numeral strings are segmented and isolated numerals are obtained using sliding window approach with user defined algorithm. Hence the present paper proposes a segmentation-recognition system using the sliding window approach with user defined classifier. The sliding window is used for discovery the interconnection spots and optimal angle for cutting the adjacent digits at the same time and a minimum of 5 features are extracted from each isolated digit for classification. The exploratory outcomes directed on a recently gathered database of manually written digits and got promising results. The overall efficiency obtained using the proposed method is about 98.51%.

  PublisherDOI

• A recursive segmentation procedure for continuous speech

  Figshare · 2018-06-30 · 10 citations

  articleOpen access

  Computer Science Department

  OA PDFDOI

• AI can help to create a humane society

  2018-05-18

  article1st authorCorresponding

  Cognition Amplifiers (COGs) and Guardian Angels (GATs) are two types of intelligent Agents technologies that assist humankind. A Cognition Amplifier is a Personal Enduring Autonomic Intelligent Agent that anticipates what you want to do and helps you to do it with less effort. A Cognition Amplifier can perform day to day tasks such as buying and selling, banking, and answer routine emails. A Guardian Angel is a Personal Enduring Autonomic Intelligent Agent assigned to each person on the planet to ensure the user's safety, security and wellbeing. A Guardian Angel can discover and warn the user about unanticipated events such as just-in-time warnings about hurricanes, earthquakes, extreme weather as well as potential impending problems of food security, water security and energy security. Together these intelligent agents can be used to create a Humane Society. AI technologies can be used to monitor, diagnose and remediate problems using personalized Guardian Angels and ensure basic necessities and protect human rights of every person on earth.

  PublisherDOI

• What a software engineer needs to know.

  Research Showcase @ Carnegie Mellon University (Carnegie Mellon University) · 2018-06-30

  articleOpen accessSenior author

  "Software development, like any complex task, requires a wide variety of knowledge and skills. We examine one particular kind of knowledge, the programming language vocabulary of the programmer, by gathering statistics on large bodies of code in three languages. This data shows that most of the identifiers in programs are either uses of built-in or standard library definitions or highly idiomatic uses of local variables. We interpret this result in light of general results on expertise and language acquisition. We conclude that tools to support the vocabulary component of software development are wanting, and this part of a engineer's education is at best haphazard, and we recommend ways to improve the situation."

  PublisherOA PDFDOI

• Rare and Frequent N-grams in Whole-genome Protein Sequences

  Figshare · 2018-06-30 · 2 citations

  articleOpen accessSenior author

  The precise relationship between a primary protein sequence, its three-dimensional structure and its function in a complex cellular environment is one of the most fundamental unanswered questions in biology. Unprecedented amounts of genomic and proteomic data create an opportunity for attacking the sequence-structure-function mapping problem with data-driven methods. The mapping of biological sequences to form and function of proteins is conceptually similar to the mapping of words to meaning. This analogy is being studied by a growing body of research ([1] and pointers thereof). Thus, n-gram analysis (statistical analysis of co-occurrence of words in a text) has found applications to biological sequences, using various types of “vocabulary”, for example nucleotides and amino acids. Here, we investigate n-gram statistics in whole-genome sequences to address the following questions: How characteristic is the amino acid n-gram distribution for specific organisms? Do different organisms tend to use different “phrases”? What is the “meaning” of a rare sequence in a protein? The long-term goal is to provide a useful starting point to derive language models with defined vocabulary and phrase preferences and grammatical rules for protein sequences of different organisms.

  OA PDFDOI

Recent grants

• Computational Learning and Discovery in Biological Sequence, Structure and Function Mapping

  NSF · $5.9M · 2002–2008

Frequent coauthors

• Judith Klein‐Seetharaman

  Arizona State University

  8 shared

• Madhavi K. Ganapathiraju

  University of Pittsburgh

  7 shared

• Philip R. Hayes

  Northumbria University

  4 shared

• Mark S. Fox

  4 shared

• Katsushi Ikeuchi

  3 shared

• Gopala K. Anumanchipalli

  2 shared

• Xuedong Huang

  2 shared

• Ömer Akın

  Istanbul Technical University

  2 shared

Awards & honors

• Fellow of the Institute of Electrical and Electronics Engine…
• Fellow of the Acoustical Society of America
• Fellow of the American Association for Artificial Intelligen…
• Member of the National Academy of Engineering
• Member of the American Academy of Arts and Sciences