Romain Thalineau

PyTorch implementation of "A Neural Algorithm of Artistic Style" from Gatys et al.

A point cloud is a set of points and can be generated by 3D scanner such as the one used by self-driving vehicles. Being unordered and irregular, features cannot be learned by simply convolving kernels against points like it is done for data represented in regular domains such as images. In this article, I expose the problem related to learning features from point clouds, review the pioneer architecture PointNet and proposes a PyTorch implementation of it.

In this notebook, I try to explain the backpropagation algorithm, heavily used in neural networks. I also proposed a bare numpy implementation of it.

As a side project, I implemented a real-time monitoring of the French presidential election on Twitter. This included:
- streaming the related tweets using the Twitter streaming API
- parsing and analyzing them
- storing them in a graph DB (Neo4J)
- updating in real time analyses
- serving these analyses via a REST API
- visualizing the analyses via a AngularJS front end

The tweet collection process is available on Github:
https://github.com/romaintha/twitter

Ph.D. project:

In this thesis we described a series of experimental works, which have been realized in the context of spin qubits, going from their use as information carriers to their use as very sensitive detectors.

We demonstrate the first experimental realization of a single electron transport along a closed path inside a system composed of four coupled quantum dots. By considering spin-orbit interaction, this experiment opens the way toward coherent topological spin manipulations. In the context of quantum computing and spin qubits, we study the two-qubit gates. By considering two tunnel-coupled quantum dots, we demonstrate by controlling the local Zeeman splitting that the natural two-qubit gate for spin qubits evolves from the SWAP gate to the C-phase gate. This work demonstrates the feasibility of the C-phase gate. Finally, we use spin qubits as very sensitive detectors. A singlet-triplet qubit is a quantum system which can be tuned in order to be very sensitive to the electrostatic environment. Here we report the use of such a qubit to detect a single electron transported next to the detector.