Quantum computing is breaking the nature’s code to deliver the promise of unprecedented growth in computational power. It enables building systems which include richer data handling and processing elements, as compared to the ones currently being used by traditional computing systems. Quantum computing reduces the training time for algorithms significantly as well as efficiently finds optimal parameters through quantum superposition and entanglement, leading to quantum parallelism. It improves predictions through Bloch sphere representation, reducing the search space for the model parameters. It also helps arrive at better solutions that work even for non-convex optimization problems.
Mphasis is at the forefront in using quantum computers for machine learning, optimization and simulation problems. As a pioneer in delivering AI/ML solutions, we foresee quantum computing as a major driver in solving our clients’ business problems.
Mphasis EON (Energy Optimized Network) is a patent pending, classical-quantum hybrid network, consisting of energy optimization, quantum circuit and deep neural network layers. A quantum assisted, deep learning architecture is the recommended approach to solve ML problems. The deep learning part handles the data preprocessing step, while the quantum hardware performs the task of parameter search. The efficiency of quantum assisted, deep learning system is much higher than the efficiency of two systems, when used separately.
Quantum machine learning works on the principle of superposition and entanglement. Feature engineering and preprocessing the data to suit the quantum properties for learning, enhances the training performance of quantum circuits, resulting in better training accuracy. Energy Optimized Network (EON) constitutes of a deep neural network and feedback loop, dedicated to learning the incoming information and transforming the data into new feature space.
Mphasis EON standardizes the classical data to bring all the values in a range of [0,1], to reduce the big number effect on machine learning. After standardization, input data is manipulated to prepare enhanced feature space that can better represent classical data on quantum systems for predictive tasks. The transformed feature space is mapped onto quantum systems using quantum state preparation such as angle embedding, amplitude embedding etc. Quantum circuits perform a defined series of operations on quantum data and measure the system information using the measurement function. The output from measurement functions becomes the input to deep neural network layers. The resultant cost function resides on the classical machine. The measured error is reduced by back propagation and different gradient descent methods.
Mphasis EON follows an iterative process to select the best suited feature space transformation, by reducing the overall energy of the network and capturing as much information spread as possible. For classical data to input on quantum computational systems, data encoding is a necessary step, and the strategy of how to represent classical data on quantum systems influence the quantum circuit design and its efficiency. A better expression of data on quantum systems, for a given machine learning task, leads to low information loss, faster convergence and better solution, due to quantum efficient optimization as compared to their classical counter parts. Mphasis EON achieves the optimal information representation for quantum using feature space transformation and batch sampling.
Strategic batch sampling methods are adopted to feed the quantum circuit such that each batch incorporates the overall pattern of information spread. This continuous and adaptive strategy to enhance feature space preparation for quantum predictive tasks make it more performance centric, resulting in faster convergence with fewer qubit requirements.
Significantly reduces the need for higher configuration quantum systems with increased data encoding efficiencies
Prepares the feature space specifically and adaptively for the quantum prediction, making the process more outcome-oriented and suitable for quantum prediction tasks
Learns data patterns and performs quantum predictive tasks in an iterative mode, enabling quantum to manage the challenging task of handling the big data effectively
Helps in better representation of information through feature space transformation by exploiting hidden patterns in data. This results in better learning of the complex patterns, thereby improving the training accuracy
Generates more orthogonal features with reduced intra-feature correlation, making it more suitable for quantum mechanism of learning the information
Reduces non-linear dimensionality of the data, making quantum circuit training feasible in less available qubits
Generates batches that guides the quantum circuit to strategically converge to optimal weights, making training faster with better accuracy
Quantum Computing allows for building of computing systems which incorporate richer data handling and processing elements than the ones currently used by traditional computers. Mphasis leverages the power of Quantum computing to solve business problems in the area of Machine Learning, Optimization and Simulation. Quantum computers help reduce the execution time for algorithms, identifies complex patterns in data to improve accuracy and help solve complex problems which are difficult to solve on classical computers in finite time. It helps improve predictions by reducing the search space for the model parameters. It also helps arrive at better solutions that works even for non-convex optimization problems. Mphasis is a pioneer in delivering AI/ML solution for clients and we foresee Quantum Computing as a major driver to support our client in solving their business problems. Mphasis offers the below Quantum Computing related professional services. These services are platform agnostic and support Quantum Annealing based systems, Quantum Circuit based systems as well as simulators and emulators.
Quantum Computing Offerings
Covid-19 related Quantum Offerings