I am a Research Scientist at DeepMind. Before joining DeepMind, I obtained my PhD from the Department of Electrical Engineering and Computer Science at MIT where
I was part of the Computer Science and Artificial Intelligence Laboratory.
During my PhD, I was advised by Tomaso Poggio in the Center for Brains, Minds, and Machines and I was part of the LCSL, a joint lab between MIT and the Istituto Italiano di Tecnologia.
Currently, I work on Relational Reasoning and I am especially interested in Hierarchical Relational Reasnoning. During my PhD I studied, and replicated in artificial systems, how human visual cortex learns to recognize objects, actions and faces in videos, and static images under a wide variety of transformations. I also design algorithms that are able to learn from large amounts of labeled data (a link to my PhD Thesis should be available soon).
I spent the summer of 2016 as a Research Intern at Google DeepMind where I worked on systems that learn Newtonian mechanics from videos. In the summer of 2014 I have worked as a Software Engineering and Research Intern at A9.com (Amazon, Inc) where I developed a Deep Learning system for object localization and recognition. In the Summer of 2013 I worked as a Data Scientist and Software Engineering Intern at Room 77, Inc (acquired by Google in 2014) where I designed and wrote algorithms for search results ranking and segmentation.
In the fall of 2013 I was a Teacher Assistant for Prof. A. Torralba's graduate class Advances in Computer Vision (6.869). In the fall of 2014 I was a Teacher Assistant for Prof. L. Kaelbling's Machine Learning (6.867) graduate class.
In 2010 I was part of the Equipment Controls and Electronic section in the Engineering Department at CERN where I developed a system to learn the minimum tracking error parameters for a complex control loop from measurements acquired on board.
During my time at MIT, I used to race bikes for the MIT Cycling Tream where I also served as Vice President. In the past I have been the organizer of the Machine Learning Tea at MIT for which I have secured fundings from Google.
Together with Sruthi Reddy Chintakunta I participated in the MIT 100K - The Entrepreneurship Competition in 2016.
- Invariant recognition drives neural representations of action sequences. Tacchetti A*, Isik L*, Poggio T. PLOS Computational Biology, 2017
- A fast, invariant representation for human action in the visual system. Isik L*, Tacchetti A*, Poggio T. Journal of Neurophysiology, 2017 (selected for APSselect, a collection from the APS that showcases some of the best recently published articles in physiological research).
- Unsupervised learning or invariant representations. Anselmi F, Leibo J, Rosasco L, Mutch J, Tacchetti A, Poggio T. Theoretical Computer Science, 2015.
- GURLS: A Least Squares Library for Supervised Learining. Tacchetti A, Mallapragada P, Santoro M, Rosasco L. Journal of Machine Learning Research 14, 3201-3205, 2013.
- Visual Interaction Networks: Learning a Physics Simulator from Video. Watters N, Tacchetti A, Weber T, Pascanu R, Battaglia P, Zoran D, 2017
- Invariant recognition drives neural representations of action sequences. Isik L*, Tacchetti A*, Poggio T. Cognitive Computational Neuroscience, 2017
- Representation Learning from Orbit Sets for One-Shot Classification. Tacchetti A*, Voinea S*, Evangelopoulos G, Poggio T. 2017 AAAI Spring Symposium Series, 2017
- Invariant representations for action recognition in the visual system. Tacchetti A*, Isik L*, Poggio, T. Journal of Vision (Oral Presentation at VSS), 2015
- Invariant representations for action recognition in the visual system. Isik L*, Tacchetti A*, Poggio, T. Computational and Systems Neuroscience (COSYNE), 2015
- Invariant representations of action recognition. Isik L*, Tacchetti A*, Poggio T. Society for Neuroscience, 2014.
- Readout of dynamic action sequences using MEG decoding. Isik L, Tacchetti A, Poggio T. Biomagnetism, 2014.
- Implementation and tuning of the Extended Kalman Filter for a sensorless drive working with arbitrary stepper motors and cable lengths. Butcher M, Masi A, Martino M, Tacchetti A. International Conference on Electrical Machines (ICEM). 2012.
- GURLS: a Toolbox for Large Scale Multiclass Learning. Tacchetti A, Mallapragada P, Santoro M, Rosasco L. NIPS Workshop on Parallel and Large-Scale Machine Learning. 2011
- Visual Interaction Networks. Watters N, Tacchetti A, Weber T, Pascanu R, Battaglia P, Zoran D. 2017
- Invariant recognition drives neural representations of action sequences. Tacchetti A*, Isik L* Poggio T. 2017
- Discriminiate-and-rectify Encoders: Learning from Image Transfromation Sets. Tacchetti A*, Voinea S*, Evangelopoulos G. 2017
- Fast, invariant representations for human actions in the visual system. Isik L*, Tacchetti A*, Poggio T. 2016
- Regularization by Early Stopping for Online Learning Algorithms. Rosasco L, Tacchetti A, Villa S. 2014 (updated version by Rosasco L, and Villa S. 2015)
- Does invariant recognition predict tuning of neurons in sensory cortex? Poggio T, Mutch J, Anselmi F, Tacchetti A, Rosasco R, Leibo J. 2013
- Unsupervised learning of invariant representations in hierarchical architectures. Anselmi F, Joel L, Rosasco L, Mutch J, Tacchetti A, Poggio T. 2013
* denotes equal contribution