-
Application of Tensorized Neural Networks for Cloud Classification
Authors:
Alifu Xiafukaiti,
Devanshu Garg,
Aruto Hosaka,
Koichi Yanagisawa,
Yuichiro Minato,
Tsuyoshi Yoshida
Abstract:
Convolutional neural networks (CNNs) have gained widespread usage across various fields such as weather forecasting, computer vision, autonomous driving, and medical image analysis due to its exceptional ability to extract spatial information, share parameters, and learn local features. However, the practical implementation and commercialization of CNNs in these domains are hindered by challenges…
▽ More
Convolutional neural networks (CNNs) have gained widespread usage across various fields such as weather forecasting, computer vision, autonomous driving, and medical image analysis due to its exceptional ability to extract spatial information, share parameters, and learn local features. However, the practical implementation and commercialization of CNNs in these domains are hindered by challenges related to model sizes, overfitting, and computational time. To address these limitations, our study proposes a groundbreaking approach that involves tensorizing the dense layers in the CNN to reduce model size and computational time. Additionally, we incorporate attention layers into the CNN and train it using Contrastive self-supervised learning to effectively classify cloud information, which is crucial for accurate weather forecasting. We elucidate the key characteristics of tensorized neural network (TNN), including the data compression rate, accuracy, and computational speed. The results indicate how TNN change their properties under the batch size setting.
△ Less
Submitted 21 March, 2024;
originally announced May 2024.
-
Quantum Phase Estimations of Benzene and Its Derivatives on GPGPU Quantum Simulators
Authors:
Yusuke Ino,
Misaki Yonekawa,
Hideto Yuzawa,
Yuichiro Minato,
Kenji Sugisaki
Abstract:
Quantum computers are expected to perform the full con-figuration interaction calculations with fewer computa-tional resources compared to classical ones, thanks to the use of the quantum phase estimation (QPE) algorithms. However, only limited number of the QPE-based quantum chemical calculations have been reported even on the numerical simulations on a classical computer, focusing on small molec…
▽ More
Quantum computers are expected to perform the full con-figuration interaction calculations with fewer computa-tional resources compared to classical ones, thanks to the use of the quantum phase estimation (QPE) algorithms. However, only limited number of the QPE-based quantum chemical calculations have been reported even on the numerical simulations on a classical computer, focusing on small molecules of up to five atoms. In this paper, we performed quantum chemical calculations of electronic ground and excited states on industrially important mole-cules using the iterative QPE algorithms. With the simula-tor based on a single-GPGPU, we observed the speedup compared to the ones based on multi-CPUs. We also con-firmed the feasibility of this method using a quantum simulator and evaluated the π-π* excitation energies of benzene and its mono-substituted derivatives. Our meth-od is easily applicable to other molecules and can be a standard approach for performing the QPE-based quan-tum chemical calculations of practical molecules.
△ Less
Submitted 26 December, 2023;
originally announced December 2023.
-
A Large-Scale Pad-Sensor Based Prototype of the Silicon Tungsten Electromagnetic Calorimeter for the Forward Direction in ALICE at LHC
Authors:
R. G. E. Barthel,
T. Chujo,
T. Hachiya,
M. Hatakeyama,
Y. Hoshi,
M. Inaba,
Y.,
Kawamura,
D. Kawana,
C. Loizides,
Y. Miake,
Y. Minato,
K. Nakagawa,
N. Novitzky,
T. Peitzmann,
M. Rossewij,
M. Shimomura,
T. Sugitate,
T. Suzuki,
K. Tadokoro,
M. Takamura,
S. Takasu,
A. van den Brink,
M. van Leeuwen
Abstract:
We constructed a large-scale electromagnetic calorimeter prototype as a part of the Forward Calorimeter upgrade project (FoCal) for the ALICE experiment at the Large Hadron Collider (LHC). The prototype, also known as ``Mini FoCal'', consists of 20 layers of silicon pad sensors and tungsten alloy plates with printed circuit boards and readout electronics. The constructed detector was tested at the…
▽ More
We constructed a large-scale electromagnetic calorimeter prototype as a part of the Forward Calorimeter upgrade project (FoCal) for the ALICE experiment at the Large Hadron Collider (LHC). The prototype, also known as ``Mini FoCal'', consists of 20 layers of silicon pad sensors and tungsten alloy plates with printed circuit boards and readout electronics. The constructed detector was tested at the test beam facility of the Super Proton Synchrotron (SPS) at CERN. We obtain an energy resolution of about 4.3% for electron beams at both 150 and 250 GeV/$c$, which is consistent with realistic detector response simulations. Longitudinal profiles of electromagnetic shower were also measured and found to agree with the simulations. The same prototype detector was installed in the ALICE experimental area about 7.5m away from the interaction point. It was used to measure inclusive electromagnetic cluster energy distributions and neutral-pion candidate invariant mass distributions for pseudo-rapidity of $η$=3.7-4.5 in proton-proton collisions at $\sqrt{s}$ = 13 TeV at LHC. The measured distributions in different $η$ regions are similar to those obtained from PYTHIA simulations.
△ Less
Submitted 18 March, 2024; v1 submitted 9 June, 2023;
originally announced June 2023.
-
Explainable Natural Language Processing with Matrix Product States
Authors:
Jirawat Tangpanitanon,
Chanatip Mangkang,
Pradeep Bhadola,
Yuichiro Minato,
Dimitris G. Angelakis,
Thiparat Chotibut
Abstract:
Despite empirical successes of recurrent neural networks (RNNs) in natural language processing (NLP), theoretical understanding of RNNs is still limited due to intrinsically complex non-linear computations. We systematically analyze RNNs' behaviors in a ubiquitous NLP task, the sentiment analysis of movie reviews, via the map** between a class of RNNs called recurrent arithmetic circuits (RACs)…
▽ More
Despite empirical successes of recurrent neural networks (RNNs) in natural language processing (NLP), theoretical understanding of RNNs is still limited due to intrinsically complex non-linear computations. We systematically analyze RNNs' behaviors in a ubiquitous NLP task, the sentiment analysis of movie reviews, via the map** between a class of RNNs called recurrent arithmetic circuits (RACs) and a matrix product state (MPS). Using the von-Neumann entanglement entropy (EE) as a proxy for information propagation, we show that single-layer RACs possess a maximum information propagation capacity, reflected by the saturation of the EE. Enlarging the bond dimension beyond the EE saturation threshold does not increase model prediction accuracies, so a minimal model that best estimates the data statistics can be inferred. Although the saturated EE is smaller than the maximum EE allowed by the area law, our minimal model still achieves ~99% training accuracies in realistic sentiment analysis data sets. Thus, low EE is not a warrant against the adoption of single-layer RACs for NLP. Contrary to a common belief that long-range information propagation is the main source of RNNs' successes, we show that single-layer RACs harness high expressiveness from the subtle interplay between the information propagation and the word vector embeddings. Our work sheds light on the phenomenology of learning in RACs, and more generally on the explainability of RNNs for NLP, using tools from many-body quantum physics.
△ Less
Submitted 6 April, 2022; v1 submitted 16 December, 2021;
originally announced December 2021.
-
Finding high-order Hadamard matrices by using quantum computers
Authors:
Andriyan Bayu Suksmono,
Yuichiro Minato
Abstract:
Solving hard problems is one of the most important issues in computing to be addressed by a quantum computer. Previously, we have shown that the H-SEARCH; which is the problem of finding a Hadamard matrix (H-matrix) among all possible binary matrices of corresponding order, is a hard problem that can be solved by a quantum computer. However, due to the limitation on the number of qubits and connec…
▽ More
Solving hard problems is one of the most important issues in computing to be addressed by a quantum computer. Previously, we have shown that the H-SEARCH; which is the problem of finding a Hadamard matrix (H-matrix) among all possible binary matrices of corresponding order, is a hard problem that can be solved by a quantum computer. However, due to the limitation on the number of qubits and connections in present day quantum processors, only low orders H-SEARCH are implementable. In this paper, we show that by adopting classical construction/search techniques of the H-matrix, we can develop new quantum computing methods to find higher order H-matrices. Especially, the Turyn-based quantum computing method can be further developed to find an arbitrarily high order H-matrix by balancing the classical and quantum resources. This method is potentially capable to find some unknown H-matrices of practical and scientific interests, where a classical computer alone cannot do because of the exponential grow of the complexity. We present some results of finding H-matrix of order more than one hundred and a prototypical experiment to find even higher order matrix by using the classical-quantum resource balancing method. Although heuristic optimizations generally only achieve approximate solutions, whereas the exact one should be determined by exhaustive listing; which is difficult to perform, in the H-SEARCH we can assure such exactness in polynomial time by checking the orthogonality of the solution. Since quantum advantage over the classical computing should have been measured by comparing the performance in solving a problem up to a definitive solution, the proposed method may lead to an alternate route for demonstrating practical quantum supremacy in the near future.
△ Less
Submitted 21 October, 2020; v1 submitted 22 September, 2020;
originally announced September 2020.
-
Solving tiling puzzles with quantum annealing
Authors:
Asa Eagle,
Takumi Kato,
Yuichiro Minato
Abstract:
To solve tiling puzzles, such as "pentomino" or "tetromino" puzzles, we need to find the correct solutions out of numerous combinations of rotations or piece locations. Solving this kind of combinatorial optimization problem is a very difficult problem in computational science, and quantum computing is expected to play an important role in this field. In this article, we propose a method and obtai…
▽ More
To solve tiling puzzles, such as "pentomino" or "tetromino" puzzles, we need to find the correct solutions out of numerous combinations of rotations or piece locations. Solving this kind of combinatorial optimization problem is a very difficult problem in computational science, and quantum computing is expected to play an important role in this field. In this article, we propose a method and obtained specific formulas to find solutions for tetromino tiling puzzles using a quantum annealer. In addition, we evaluated these formulas using a simulator and using actual hardware DW2000Q.
△ Less
Submitted 3 April, 2019;
originally announced April 2019.
-
Finding Hadamard matrices by a quantum annealing machine
Authors:
Andriyan Bayu Suksmono,
Yuichiro Minato
Abstract:
Finding a Hadamard matrix (H-matrix) among the set of all binary matrices of corresponding order is a hard problem, which potentially can be solved by quantum computing. We propose a method to formulate the Hamiltonian of finding H-matrix problem and address its implementation limitation on existing quantum annealing machine (QAM) that allows up to quadratic terms, whereas the problem naturally in…
▽ More
Finding a Hadamard matrix (H-matrix) among the set of all binary matrices of corresponding order is a hard problem, which potentially can be solved by quantum computing. We propose a method to formulate the Hamiltonian of finding H-matrix problem and address its implementation limitation on existing quantum annealing machine (QAM) that allows up to quadratic terms, whereas the problem naturally introduces higher order ones. For an M-order H-matrix, such a limitation increases the number of variables from M^2 to (M^3 +M^2-M)/2, which makes the formulation of the Hamiltonian too exhaustive to do by hand. We use symbolic computing techniques to manage this problem. Three related cases are discussed: (1) finding N < M orthogonal binary vectors, (2) finding M-orthogonal binary vectors, which is equivalent to finding a H-matrix, and (3) finding N-deleted vectors of an M-order H-matrix. Solutions of the problems by a 2-body simulated annealing software and by an actual quantum annealing hardware are also discussed.
△ Less
Submitted 26 March, 2019; v1 submitted 21 February, 2019;
originally announced February 2019.
