Andreas Zinonos

Andreas Zinonos

I'm a Machine Learning Researcher & Engineer interested in multimodal AI, computer vision, generative image/video, representation learning, and large language models. I recently completed my PhD at Imperial College London, supervised by Maja Pantic, where I researched deep representation learning for audio-visual speech recognition and synthesis, with publications at CVPR and ICASSP. I also hold an MSc in AI & Machine Learning from Imperial and a BSc in Computer Science from UCL. My research experience spans Meta and Cantina Labs, with additional industry experience at Qube Research & Technologies and Credit Suisse.

Outside of research, I'm an ex-filmmaker, play the guitar, and enjoy martial arts and running.

LinkedIn GitHub YouTube Scholar

Publications

CVPR 2026
FlashLips: 100-FPS Mask-Free Latent Lip-Sync using Reconstruction Instead of Diffusion or GANs
Zinonos A.*, Stypułkowski M., Bigata A., Petridis S., Pantic M., Drobyshev N.* (*equal contribution)
A two-stage, mask-free generative lip-synchronization system achieving 100+ FPS with near diffusion-level quality in perceptual metrics and human evaluation.
ICASSP 2024
BRAVEn: Improving Self-Supervised Pre-Training for Visual and Auditory Speech Recognition
Haliassos A.*, Zinonos A.*, Petridis S., Pantic M. (*equal contribution)
State-of-the-art in both low- and high-resource settings, including 20.0%/1.7% WER on VSR/ASR using only 30 hours of labels.
ICASSP 2023
Learning Cross-Lingual Visual Speech Representations
Zinonos A., Haliassos A., Ma P., Petridis S., Pantic M.
Shows that multilingual pre-training boosts low-resource visual speech recognition under monolingual video-only fine-tuning.

Research Experience

Apr 25 – Oct 25
Research Scientist Intern
Cantina Labs
London, UK
Joint-first author of FlashLips (CVPR 2026), a mask-free latent video generation system for lip-sync — best FID, FVD and LipScore at 100+ FPS, up to 30× faster than prior methods. Built spatio-temporal image and video generation models in a VAE latent space, with flow matching and custom inference strategies. Training optimization on 8× H100 GPUs: profiling, model compilation, FlashAttention 3, mixed precision.
May 24 – Nov 24; Jan 25 – Mar 25
Research Scientist Intern
Meta
London, UK
Generated and processed synthetic video datasets (up to 433 hours each) with audio- and image-conditioned diffusion and GAN models, then benchmarked synthetic against real data across pre-training, fine-tuning and testing. Built self-supervised ResNet CNN and transformer pipelines for audio-visual identity–content disentanglement, with distributed multi-GPU training.
Sep 21 – Jul 23
AI Research Analyst — Contract
Meta
London, UK
Co-developed BRAVEn (ICASSP 2024), a self-supervised audio-visual representation learning method reaching state of the art with only 30 hours of labels: 20.0%/1.7% WER on visual/audio speech recognition. Large-scale distributed training with PyTorch DDP on AWS Slurm clusters (32× A100 GPUs); built BYOL pipelines for image, audio and video representation learning and explored VICReg extensions. Self-supervised cross-lingual transfer improved low-resource visual speech recognition by up to 25% (ICASSP 2023).

Additional Work Experience

Oct 25 – Mar 26
Quantitative Research Intern
Qube Research & Technologies
Dubai, UAE
Built machine learning pipelines for predictive signal discovery on time series, with mathematics- and statistics-based feature engineering and walk-forward out-of-sample validation.
Aug 19 – Sep 21
Data Scientist & Software Developer
Credit Suisse
London, UK
Developed and deployed unsupervised machine learning for anomalous payment detection, supporting production risk monitoring across thousands of transactions. Prototyped NLP-based systems for large-scale IT ticket clustering. Designed low-latency data caching pipelines for trading systems, and built automated headless testing infrastructure for critical web applications.

Education

2021 – 2026
PhD in AI & Machine Learning
Imperial College London
Thesis: Deep Representation Learning for Audio-Visual Speech Recognition and Synthesis. Supervised by Maja Pantic. 3 first-author publications (CVPR, ICASSP ×2).
2018 – 2019
MSc in AI & Machine Learning Distinction
Imperial College London
Thesis: Predicting Brain Glioma Growth using Generative Adversarial Networks.
2015 – 2018
BSc in Computer Science First Class
University College London
Thesis: Catena AR – iOS App for Virtual Production in Film.

Awards

2025Ranked 25th / 849 in QRT Stock Prediction Challenge
20191st Place, Imperial College Data Science Insurance Pricing Competition
2018IBM Joint-First, Best Undergraduate Final Year Project
2006-2011Four National Math Olympiad Medals (Silver: 2006/2008/2011; Gold: 2007), Cyprus Mathematical Society

Talks & Workshops

Self-Supervised Learning — Internal Technical Talk, Meta, London (Jul 2024)
Self-Supervised Learning — Lecture & Workshop, AI Tech School, Warsaw (Jun 2023)
Cross-Lingual Visual Speech Representations — Poster, ICASSP 2023, Rhodes