Sai Prasath Suresh Email & Phone Number

Sunnyvale, Ca, Us

• Building RAG and function calling agents for cybersecurity

Atlanta, Georgia, United States

• Designing defense techniques for improving the robustness of BERT, GPT and Llama2 LLMs against adversarial attacks and malicious inputs. • Evaluating performance across 3 settings: prompt engineering (few shot learning), prompt tuning, and LoRA.• Building guardrails leveraging LLM embeddings to effectively identify and prevent the generation of harmful (toxic or bias) content.

Atlanta, Georgia, United States

• Enhanced the ML Testing toolbox by integrating: feature visualizations based on mutual information, and explainable AI (LIME and SHAP) for multi-class classification problems.• Evaluated the end-to-end development of machine learning models: (1) CNN for 5G Beam Selection, and (2) Autoencoder Based Channel Estimator using Keysight’s AI-testing pipeline.• Contributed to the AI Testing White Papers on Supervised and Unsupervised learning models for 6G Technology

Atlanta, Georgia, United States

• Developing a Emergency Management portal using Flask for detecting flooding in Georgia counties.• Designing pipelines for processing and storing data in GCP collected from sensors across the Georgia coast.• Building a dashboard to visualize real-time data which will be used for emergency planning and response.

Bhubaneswar, Odisha, India

Research Focus: Deep Learning, Network Security, Reinforcement LearningProject 1: Multi-Class Attack Classification using Reinforcement Learning• Designed a Double Deep Q Network (DDQN) which dynamically adapts to changes in data distribution using feedback from the user. Achieved +6% accuracy compared to existing benchmarks.• Investigated the performance of the model on the network data from the cloud based ISOT-CID (8Tb) dataset.Project 2: Covariate Shift… Show more Research Focus: Deep Learning, Network Security, Reinforcement LearningProject 1: Multi-Class Attack Classification using Reinforcement Learning• Designed a Double Deep Q Network (DDQN) which dynamically adapts to changes in data distribution using feedback from the user. Achieved +6% accuracy compared to existing benchmarks.• Investigated the performance of the model on the network data from the cloud based ISOT-CID (8Tb) dataset.Project 2: Covariate Shift Mitigation using Continual Learning• Identified the problem of covariate shift in the NSL-KDD dataset and proposed an seven-stage framework to detect the nature and magnitude of the change.• Extended replay(dark experience replay) and regularization(learning without forgetting) based continual learning models from computer vision to intrusion detection tasks to handle changing attack distributions. Show less

Singapore

• Developed a novel semi-supervised GAN model for detecting trojan DNNs created by service providers.• Redesigned existing models to achieve attack-agnostic detection by integrating an Denoising Autoencoder.• Achieved state-of-the-art performance +3% AUC on computer vision tasks while minimizing run-time by 15%.

Singapore

• Implemented a dual attention based LSTM/GRU models to pre-emptively detect anomalies in a power plant.• Minimized costs by reducing the false alarm rates to 0.21% with a high detection accuracy of 97.8%.• Compared the performance of the model with ML models(Random Forests, XGBoost, SVM) and DNNs.

Karlsruhe, Baden-Württemberg, Germany

• Redesigned multiple adversarial attacks such as Zeroth Order Optimization(ZOO) and Iterative Fast Gradient Sign Method (FGSM) to accommodate domain specific needs.• Evaluated various defence mechanisms including denoising autoencoder, adversarial training, and defensive distillation.

Auckland, New Zealand

• Implemented agent(Stochastic Te Punaha Matatini) and differential equation(Modified SEIR, CovidSim 2.0) based covid-19 models using hybrid automata using composition modelling.• Analysed the trade off between the economic impact and the health impact by generating a pareto front using ϵ constraint method and particle swarm optimization.

Master'S Degree, Computer Science (Specializing In Machine Learning), 4.0/4.0

Course Work: Fall 2022 Machine Learning (Prof. Mahdi Roozbahani) Interactive Robot Learning (Prof. Matthew Gombolay) Computation Science.

Bachelor Of Technology, Computer Science And Engineering, 9.79/10

Activities and Societies: 1. Secretary of the Literary Society, Panacea 2. Assistant Coordinator of the Counselling Service Team 3. Core.

Class 12, Computer Science, Maths, Physics, Chemistry, 95.8%

Class 10, Maths, Science, Social, English

CGPA: 10/10