Loris Bazzani

Bio

I am an AI Research Leader with 15+ years of experience advancing AI, from classical computer vision and machine learning to today’s foundation and multimodal generative models. My core strength lies in driving innovation: developing new models that push the state of the art and translating them into real-world products at global scale.

As Principal Scientist at Amazon, I led high-impact core research and product efforts across Prime Video, Alexa, and mobile/.com shopping. I co-developed novel models and architectures for video understanding, vision-language representation, Large Multimodal Models (LMMs), and diffusion models. My work powered AI-driven features such as live sports highlights, virtual try-on, interactive product recommendations, and shopping assistants, used by millions of users worldwide and generating O(XXM) USD in business impact. I pioneered Amazon’s first deep learning models for video understanding and LMM-based shopping experiences, laying the foundation for new product directions and research programs across the company.

My research has been published in top-tier venues including CVPR, ICCV, ECCV, and ICML, with contributions spanning vision-language training, video understanding, generative models, and LMMs. I’ve authored 50+ publications and granted patents, and actively serve the scientific community as a reviewer and workshop organizer.

I thrive in fast-paced, cross-functional environments, collaborating with scientists, engineers, and product leaders to turn foundational research into scalable solutions. My goal is to build AI systems that are both technically groundbreaking and product-relevant.

I obtained my Ph.D. in Computer Science from the University of Verona (Italy) in 2012 supervised by Prof. Vittorio Murino and Prof. Marco Cristani. During my Ph.D., I visited the University of British Columbia collaborating with Prof. Nando de Freitas. Before Amazon, I was a postdoctoral fellow at Dartmouth College working with Prof. Lorenzo Torresani and I was a postdoctoral fellow at the Italian Institute of Technology working with Prof. Vittorio Murino.

Research and Publications [Google Scholar]

Learning Visual Hierarchies in Hyperbolic Space for Image Retrieval

Z. Wang, S. Ramasinghe, C. Xu, J. Monteil, L. Bazzani, T. Ajanthan

In International Conference on Computer Vision (ICCV), 2025

Loris Bazzani

Bio

Research and Publications [Google Scholar]

Learning Visual Hierarchies in Hyperbolic Space for Image Retrieval

LatteCLIP: Unsupervised CLIP Fine-Tuning via LMM-Synthetic Texts

ViewFusion: Towards Multi-View Consistency via Interpolated Denoising

iEdit: Localised Text-guided Image Editing with Weak Supervision

Contrastive Language-Action Pre-training for Temporal Localization

Learning Attribute-driven Disentangled Representations for Interactive Fashion Retrieval

Revamping Cross-Modal Recipe Retrieval with Hierarchical Transformers and Self-supervised Learning

Localized Triplet Loss for Fine-Grained Fashion Image Retrieval

Learning Joint Visual Semantic Matching Embeddings for Language-guided Retrieval

Image Search with Text Feedback by Visiolinguistic Attention Learning

Image Captioning as Neural Machine Translation Task in SOCKEYE

Recurrent Mixture Density Network for Spatiotemporal Visual Attention

Group Detection and Tracking using Sociological Features

Approximate Log-Hilbert-Schmidt distances between covariance operators for image classification

Self-taught object localization with deep networks

A Unifying Framework in Vector-valued Reproducing Kernel Hilbert Spaces for Manifold Regularization and Co-Regularized Multi-view Learning

Kernel Methods on Approximate Infinite-Dimensional Covariance Operators for Image Classification

Joint Individual-Group Modeling for Tracking

SDALF: modeling human appearance with symmetry-driven accumulation of local features

Weighted bag of visual words for object recognition

A unifying framework for vector-valued manifold regularization and multi-view learning

Semi-supervised multi-feature learning for person re-identification

Person re-identification with a PTZ camera: an introductory study

Symmetry-driven accumulation of local features for human characterization and re-identification

Social interactions by visual focus of attention in a three-dimensional environment

Decentralized particle filter for joint individual-group tracking

Learning where to attend with deep architectures for image tracking

Re-identification with RGB-D sensors

Online bayesian non-parametrics for social group detection

Analyzing groups: a social signaling perspective

Multiple-shot person re-identification by chromatic and epitomic analyses

Learning attentional policies for object tracking and recognition in video with deep networks

Custom pictorial structures for re-identification

Social interaction discovery by statistical analysis of F-formations

Towards computational proxemics: Inferring social relations from interpersonal distances

Multiple-shot person re-identification by hpe signature

Person re-identification by symmetry-driven accumulation of local features

Collaborative particle filters for group tracking

Timeline

Principal Scientist, Amazon

Postdoc, Dartmouth College

Postdoc, Italian Institute of Technology

PhD in Computer Vision, University of Verona

Visiting Student, University of British Columbia

M.S., University of Verona

B.S., University of Verona

Theses

Beyond Multi-target tracking: statistical pattern analysis of people and groups

Particle filtering approaches for multi-target tracking in video surveillance applications

Techniques for the Analysis and the Classification of MRI for Searching Pathology with Application to Mental Health