Harshitha Machiraju

AI Researcher · PhD in Machine Learning, EPFL

harshitha.acad@gmail.com github.com/harsmac linkedin.com/in/harshitha-machiraju Google Scholar

Download PDF

AI researcher with a PhD in Machine Learning from EPFL, working on adversarial robustness, distribution shift, image corruptions and bias mitigation. I currently red-team LLM-as-judge reward models as a LASR Labs Fellow, mentor on AI safety at BlueDot Impact, and build LLM-based systems independently. My work spans computer vision, NLP, multi-modal learning and LLM safety.

Experience

Jul 2026 – Present

LASR Labs Fellow

London, UK

Red-teaming LLM-as-judge reward models through RL and steering. Supervised by Andrew Draganov.

Apr 2026 – Present

AI Safety Expert & Mentor

BlueDot Impact · Remote

Working on AI alignment, robustness, frontier AI risks, and LLM safety evaluations.

Jan 2025 – Present

Independent AI Engineer

Remote

Designed and trained scalable LLM-based systems for health diagnostics and personalized shopping startups.

Sep 2019 – Nov 2024

PhD in Machine Learning

EPFL, Switzerland

Designed and implemented robust AI models for various applications, focusing on efficiency and resilience to distribution shifts.

Sep 2018 – Aug 2019

Research Assistant

IIT Hyderabad, India

Developed and deployed ML models for autonomous navigation, including the implementation of adversarial testing frameworks.

Research Highlights

Work not covered by the publications below. Longer write-ups live on the projects page.

Adversarial Subspace Analysis in LLMs

Developed a method to identify low-dimensional subspaces within word embeddings that concentrate the most discriminative features. Demonstrated that critical information learned by LLMs is often compactly represented in these subspaces.

Fairness vs. Adversarial Robustness in LLMs

Demonstrated that adversarial robustness does not guarantee fairness, revealing persistent biases in robust LLMs and highlighting the need for comprehensive fairness evaluations.

MUFIA: automating out-of-distribution sample generation

An algorithm that automates the generation of out-of-distribution samples by harnessing the spectral biases of models, using those biases to produce adversarial image corruptions.

Selected Publications

  1. H. Machiraju, M. Herzog, P. Frossard. “EREN: Enhancing deep learning robustness through image pre-processing.” Under review, 2024
  2. H. Machiraju, M. Herzog, P. Frossard. “Frequency-based vulnerability analysis of deep learning models against image corruptions.” Under review, 2023
  3. H. Machiraju, O. Choung, M. Herzog, P. Frossard. “Empirical advocacy of bio-inspired models for robust image recognition.” CVPR NeuroVision Workshop, 2022
  4. K. Wang, H. Machiraju, O. Choung, M. Herzog, P. Frossard. “CLAD: A contrastive learning based approach for background debiasing.” BMVC, 2022
  5. H. Machiraju, V. Balasubramanian. “A Little Fog for a Large Turn.” WACV, 2020
  6. N. Kumari, M. Singh, A. Sinha, H. Machiraju, B. Krishnamurthy, V. Balasubramanian. “Harnessing the vulnerability of latent layers in adversarially trained models.” IJCAI, 2019
  7. H. Machiraju, S. Channappayya. “An evaluation metric for object detection algorithms in autonomous navigation systems and its application to a real-time alerting system.” ICIP, 2018 (oral)

Complete list on Google Scholar.

Education

Sep 2019 – Nov 2024

PhD in Machine Learning

EPFL, Switzerland

Thesis: Towards Robust Vision Models. Advisors: Prof. Pascal Frossard & Prof. Michael Herzog.

Jul 2014 – Aug 2018

B.Tech in Electrical Engineering

IIT Hyderabad, India

Summa cum laude, with a minor in Computer Science.

Skills

Programming
Python, C, C++, Java, Matlab, SQL, Kubernetes, Docker, Slurm
Frameworks
PyTorch, TensorFlow, LangChain, Weights & Biases, Hugging Face, Git, LaTeX, Illustrator
Languages
English (native), Hindi (native), Telugu (native), Korean (intermediate), French (basic)
Certifications
BlueDot Impact: Transformative AI, AI Alignment, AI Governance

Awards & Service

Awards and recognition

  • DeepVision Grant, 2019–2021
  • Academic Excellence Award, IIT Hyderabad, 2014
  • JICA Scholarship 2018, JENESYS Scholarship 2017, KVPY 2013
  • Special Recognition for a Young Team, IEEE SP Cup, 2016
  • Top 10 teams, IEEE SP Cup, 2016

Community service

  • Reviewer for CVPR, ECML, TIP, ICVGIP
  • Supervision of Master's student projects
  • Teaching assistant, Signal Processing & Deep Learning