Brief description of the current research

Currently, I am interested in developing classical and quantum optical artificial intelligence systems such as optical deep neural networks, optical reservoir computing devices, optical Ising machines, and Quantum Computing. These smart optical techniques will provide high computational speed for wide range of complex problems not easily provided by existing von Neumann computing architectures. Some of these problems are Non-deterministic Polynomial time hard (NP-hard) to solve using commercially available computers. Optical Deep Neural Networks:Recently, deep/convolutional neural networks have been hot topics in machine artificial intelligence. These machine learning methods are extreme fast to solve problems like image reconstruction, machine translate, speech recognition and image classification. We propose a novel optical feature extraction method based on nonlinear sum frequency (SF) generation process which upconverts input images using structured Laguerre-Gaussian (LG) modes. The detected features then are used in deep neural networks (DNNs)to classify the input patterns. We show that this nonlinear mode-selective method can extract the features of MNIST handwritten digits with no more than 40 Laguerre-Gaussian pump modes using single pixel imaging, which can speedup the pattern recognition task for large size high-resolution images.Optical Reservoir Computing Devices:Unlike recurrent neural network (RNN), a typical Reservoir Computing (RC) device required training only at the output layer. The RC consists of 3 layers: input, reservoir, and output. Recently, the RC based on a single dynamical node has been proposed. Such method is capable of integrating large number of virtual nodes with limited elements and utilizing fast processing and complex properties of optics, achieved, e.g., by using electro-optical modulation for nonlinear nodes. We experimentally demonstrate a fully-packaged electro-optic RC system consisting of an electro-optical modulator and a field programmable gate array (FPGA). To characterize the reservoir computer, we demonstrate the performance of NARMA10 benchmark task by evaluate the normalized root mean squared error(NRMSE). We show that minimum NRMSE of the free prediction is reaching record low 0.134and matched with the simulation result. Optical Ising Machine:Finding the optimum solution of NP-hard problems could be beneficial for many branches of science and technology. Many of these problems can be mapped to the Ising Hamiltonian problem such as MAX-CUT, protein folding, traveling sales-man, and so on. To find the optimum solutions of these problems, optical Ising machines have the advantages of high-speed and connectivity. For instance, a coherent Ising machine can be realized with temporally multiplexed pulses using an optical parametric oscillator. However, this method is limited with spin numbers and relying on electronic feedback to emulate spin-spin interactions. In contrast, a linear-optical Ising machine based on spatial light modulation (SLM) can subtend to millions-of-spins and spin-spin interaction is emulated on the far-field using Fourier lens. This method has scalability and all-to-all connectivity which make it an attractive approach to find the optimum solution of fully connected two-body interaction. Yet, there are physical systems whose dynamics cannot be captured by two-body interactions, and proper descriptions of multi-body interaction are required. A small class of many-body interactions