Avatar

Sajjad Abdoli

Ph.D. Founding AI Scientist

Perle.ai

Biography

Hi, I’m Sajjad Abdoli, a Founding AI Scientist at Perle.ai, a San Francisco–based startup dedicated to bringing greater humanity, transparency, and reliability to artificial intelligence through expert-in-the-loop systems. My work focuses on building robust, evaluable, and safe AI technologies that align closely with real human needs and expertise.

Before Perle, I was a Machine Learning Scientist at Ubenwa health, a Mila - Quebec AI Institute affiliated startup in Montreal developing machine-learning-driven tools to uncover neurological and physiological insights from infant cry analysis.

I completed my PhD in Machine Learning in 2022 at École de technologie supérieure in Montreal, Canada. My research interests span a wide range of AI domains, including large language models (LLMs), model evaluation and benchmarking, audio and speech processing, visual object recognition and localization, and the security and safety of machine learning systems.

Outside of research, I’m passionate about running, hiking, and mobile photography

Interests

  • LLMs, LLM evals
  • VLMs
  • Audio and Speech Processing
  • Music Information Retrieval
  • Adversarial Machine Learning
  • Object localization and classification

Education

  • PhD in Artificial Intelligence, 2017 - 2022

    École de technologie supérieure, Université du Québec

  • MEng in Computer Engineering, 2017

    Azad University, Qazvin Branch

  • BSc in Computer Engineering, 2013

    Azad University, Central Tehran Branch

Publications

Quickly discover relevant content by filtering publications.

End-to-end environmental sound classification using a 1D convolutional neural network

We present an end-to-end approach for environmental sound classification based on a 1D Convolution Neural Network (CNN) that learns a …

Universal Adversarial Audio Perturbations

We demonstrate the existence of universal adversarial perturbations, which can fool a family of audio classification architectures, for …

Iranian Traditional Music Dastgah Classification

In this study, a system for Iranian traditional music Dastgah classification is presented.

Talks

Speaker Detection in adverse scenarios with single microphone Project

Closing ceremony final presentation at JSALT 2019 workshop.