QuantHD: A Quantization Framework for FPGA Acceleration of Hyperdimensional Computing

• Journal: IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems (TCAD)
• M. Imani, S. Bosch, S. Datta, S. Ramakrishna, S. Salamat, J. M. Rabaey, T. Š. Rosing
• Research done at University of California, San Diego
• Contribution: QuantHD is a Machine Learning algorithm which can achieve very significant improvements in terms of energy-efficiency and speed, without losing classification accuracy as compared to low-cost state-of-the-art Machine Learning algorithms
• My contribution: I developed large parts of the algorithm, coded the entire software part, developed the underlying mathematics and wrote parts of the paper
• Topic of my TEDx talk at EPFL: go.epfl.ch/TED

SemiHD: Semi-Supervised Learning Using Hyperdimensional Computing

• Conference: IEEE/ACM International Conference On Computer Aided Design (ICCAD)
• M. Imani, S. Bosch, M. Javaheripi, B. Rouhani, X. Wu, F. Koushanfar, T. Š. Rosing
• Research done at University of California, San Diego
• Contribution: We developed a new semi-supervised Machine Learning algorithm. We noticed a significant increase in performance, as compared to state-of-the-art algorithms.
• My contribution: I was responsible for coding the entire project, improving the mathematical algorithm and testing it against other state-of-the-art algorithms

AdaptHD: Adaptive Efficient Training for Brain-Inspired Hyperdimensional Computing

• Conference: IEEE Biomedical Circuits and Systems (BioCAS)
• M. Imani, J. Morris, S. Bosch, H. Shu, G. De Micheli, T. Š. Rosing
• Research done at University of California, San Diego and EPFL, Lausanne
• Contribution: Improvement to previous Hyperdimensional computing-based Machine Learning algorithms by introducing adaptive learning rates. We managed to significantly decrease the convergence time during training.
• My contribution: I was the first person to introduce the concept of learning rates in HD computing and performed the initial tests with adaptive learning rates. My colleagues at UCSD developed the rest of the project and wrote the paper.

CompHD: Efficient Hyperdimensional Computing Using Model Compression

• Conference: IEEE/ACM International Symposium on Low Power  Electronics and Design (ISLPED)
• J. W. Morris, M. Imani, S. Bosch, B. L. Baxter, H. Shu, T. Š. Rosing
• Research done at University of California, San Diego
• Contribution: We invented an algorithm for ultra-low-power devices with limited memory to perform offline Machine Learning. It does so by compressing multiple vectors into just one vector by encrypting them with randomized vectors, and then decrypting them during inference.
• My contribution: I was responsible for coding most of the project and improving the mathematics and statistics behind the algorithm. I invented an orthogonalization method, which significantly improved the classification accuracy by decreasing the influence of noise caused by the randomness.

Particle shape accounts for instrumental discrepancy in ice core dust size distributions

• Journal: Climate of the Past / European Geosciences Union
• M. F. Simonsen, L. Cremonesi, G. Baccolo, S. Bosch, B. Delmonte, T. Erhardt, H. A. Kjær, M. Potenza, A. Svensson, P. Vallelonga
• Research done at the Niels Bohr Institute, University of Copenhagen
• Contribution: Significant progress in measurement techniques for ice and climate science.
• My contribution: I was the first person to have had a detailed look into the dust particle size distribution from ice cores from the North Pole. I found out that laser measurement techniques used for the past 20 years have been very inaccurate. My colleagues theoretically and experimentally confirmed this and wrote the paper on it.

Quantum polar decomposition algorithm

• arXiv:2006.00841
• Submitted to the PRL
• S. Lloyd, S. Bosch, G. De Palma, B. Kiani, Z. Liu, M. Marvian, P. Rebentrost, D. M. Arvidsson-Shukur
• Research done at MIT & Harvard
• Contribution: We developed a quantum polar decomposition algorithm, which can be used for performing quantum machine learning on quantum data (learning unitaries)

A Stochastic Acceleration Method for HD Computing-Based Machine Learning

• arXiv:1911.12446
• Submitted to the IEEE Design Automation Conference (DAC)
• S. Bosch, A. Sanchez de la Cerda, M. Imani, T. Š. Rosing, G. De Micheli
• Research done at EPFL
• Contribution: We developed a new stochastic retraining method for improving the classification accuracy of HD computing-based Machine Learning algorithms which decreases the convergence rate during training by 50% and improves the overall classification accuracy of the algorithm.
• My contribution: I invented the algorithm, developed the theory and wrote the paper. The algorithm was implemented and evaluated by my colleagues at EPFL and UCSD.

Bell Diagonal and Werner state generation: entanglement, non-locality, steering and discord on the IBM quantum computer

• arXiv:1912.06105
• E. Gårding, N. Schwaller, S. Chang, S. Bosch, W. R. Laborde, J. N. Hernandez, C. L. Chan, F. Gessler, X. Si, M. A. Dupertuis, N. MacrisResearch done at EPFL
• Contribution: We developed efficient circuits for measuring entanglement and discord of quantum states and implemented them on IBM quantum computers.
• My contribution: I co-organized the research project with the EPFL Quantum Computing Association, wrote two sections of the paper and helped with the theory and implementation on real IBM quantum computers.

Convex Optimization Algorithms for Quantum Computers with Applications in Computational Finance

• P. Rebentrost, S. Bosch, S. Lloyd
• Research done at CQT, NUS and MIT
• Contribution: We exhibit a quantum algorithm for solving convex optimization problems by exploiting game theory and quantum chemistry algorithms.
• My contribution: I helped developing parts of the algorithm and tested it. Also, I worked on the underlying mathematical theory and helped with the applications in finance.
• I presented the preliminary results in the IWQC workshop at the ICCAD conference 2019.
• Topic of my seminar talk at Stanford University (Gates CS building) in November 2019 (hosted by Prof. Subhasish Mitra).

Global increase in atmospheric dust content over the last 300 years

• S. Bosch, M. F. Simonsen, H. A. Kjær, A. Svensson, P. Vallelonga
• Research done at the Niels Bohr Institute, University of Copenhagen
• Contribution: Discovery of a large increase in atmospheric dust content, which will have significant impact on our climate.
• My contribution: I made the discovery, analyzed all the data and wrote the entire paper. We are waiting with the submission to Nature Climate Change, as we still need more measurements from the North Pole, which cannot be done before April/May 2020.