Search and agents
Retrieval, evaluation, tool use, and agent experiences grounded in useful information.
AI researcher · Applied scientist
Andrea Maracani
Senior Applied Scientist at Microsoft
I am a machine learning researcher working on search, Copilot, and agents. My research concerns the efficiency, robustness, and scalability of modern AI systems, from selecting and curating data to training, evaluation, and useful deployed experiences.
Research
Research interests
My work studies how AI systems can make better use of data and computation while remaining reliable under practical constraints.
Efficient learning
Data selection, pre-training, knowledge transfer, distillation, and efficient architectures.
Multimodal intelligence
Models that connect language and vision for recognition, generation, and reasoning.
Robust adaptation
Continual learning, domain adaptation, post-training, and behavior under distribution shift.
Research record
Publications
2026
-
ECCV 2026
Geometrically Consistent Multi-View Scene Generation from Freehand Sketches
-
ACL Industry 2026
Decoding Text Spans for Efficient and Accurate Named-Entity Recognition
-
ICASSP 2026
Improving Cross-Domain Generalization of Lightweight Transformers on Named Entity Recognition Using Self-Training
2025
-
CVPR 2025
Accurate Scene Text Recognition with Efficient Model Scaling and Cloze Self-Distillation
-
ACM MMSys 2025
Efficient and Accurate Scene Text Recognition with Cascaded-Transformers
-
IEEE Sensors Journal 2025
Sim2Surf: A Sim2Real Surface Classifier for Vision-Based Tactile Sensors
-
EMNLP Industry 2025
Multi-Task Pre-Finetuning of Lightweight Transformer Encoders for Text Classification and NER
2024
-
Preprint 2024
Trust And Balance: Few Trusted Samples Pseudo-Labeling and Temperature Scaled Loss for Effective Source-Free UDA
-
Preprint 2024
Key Design Choices in Source-Free Unsupervised Domain Adaptation: An In-depth Empirical Analysis
-
ICRA 2024
Sim2Real Bilevel Adaptation for Object Surface Classification using Vision-Based Tactile Sensors
2020—2023
-
Scientific Reports 2023
In-domain versus out-of-domain transfer learning in plankton image classification
-
Preprint 2023
Key Design Choices for Double-Transfer in Source-Free Unsupervised Domain Adaptation
-
IEEE RO-MAN 2022
From Handheld to Unconstrained Object Detection: a Weakly-supervised On-line Learning Approach
-
ICCV 2021
RECALL: Replay-Based Continual Learning in Semantic Segmentation
-
Technologies 2020
Unsupervised Domain Adaptation in Semantic Segmentation: a Review
Professional background
Experience
-
2026 — present
Senior Applied Scientist
Microsoft · London, United Kingdom
Search for Copilot and agent systems.
-
2024 — 2026
Senior Researcher
Samsung Research · London, United Kingdom
Research across natural language processing, computer vision, and multimodal learning.
-
2023 — 2024
Research Intern
Amazon Web Services · Tübingen, Germany
Data selection and pruning for large language model pre-training.
-
2022 — 2023
Research Intern
Amazon Prime Air · Graz, Austria
Annotation-efficient semantic segmentation in high-resolution imagery.
-
2019 — 2020
Research Intern
University of Padua · Padua, Italy
Continual learning for semantic segmentation with generative replay.
Academic background
Education
-
2020 — 2024
PhD in Machine Learning and Computer Vision
Istituto Italiano di Tecnologia and University of Genoa
Knowledge Transferability for Data-Efficient Deep Learning
-
2017 — 2020
MSc in Information Engineering
University of Padua
Incremental Learning in Semantic Segmentation with Generative Replay Models
-
2013 — 2017
BSc in Information Engineering
University of Padua
Handover Prediction on Wi-Fi Systems with Neural Networks