Research Thrusts

Neural networks are frameworks for learning that take inspiration from biological systems and are commonly used in applications related to artificial intelligence and machine learning. Reservoir computing is a type of deep learning framework based on recurrent neural networks (RNNs) that utilizes randomly connected neurons with untrained internal layers. This architecture is both fast and efficient, with a less expensive training process compared to traditional RNNs. The reservoir can be thought of as a complex nonlinear dynamic filter that maps low-dimensional inputs into a higher dimensional space. However, despite its advantages, reservoir-based RNNs have received relatively little attention compared to other types of RNNs. As part of my research, I aim to explore the design and optimization of such RNNs to enhance their potential for use in machine learning, information processing, controller design, natural language processing (NLP), and quantum computation. To achieve, I use nonlinear physical systems and exploit the physics/dynamics of such systems in building the reservoir (aka neural net) for computation. Such approach of using a physical system as a reservoir is known as Physical Reservoir Computing. By utilizing the physical system as a reservoir, its dynamics directly participate in computation, eliminating the distinction between dynamics and computation. 

Towards this direction, my long term goal is to understand and mimic the human brain more closely to optimize the current neuromorphic architectures and apply them in artificially intelligent devices. Kindly refer to the publication page to review the research endeavors in this field. 

Whereas linear oscillators have static natural frequencies, adaptive oscillators (AOs) are a type of nonlinear oscillator that can both learn and store information in its plastic states. For instance, adaptive frequency oscillators (AFOs) are a subset of adaptive oscillators, which contain a single plastic state that enables it to learn and store a frequency component from an external forcing. Compared to non-adaptive oscillators, adaptive oscillators have received relatively little attention. Little work has been completed to analyze their dynamics and to implement them as physical experiments. In literature, such oscillators have been proposed to detect gait phase and frequency in robots for assisting people with walking difficulties due to muscle weakness.

Towards this direction, my goal is to bolster research and technology in the areas of hardware signal processing, information processing, medical implants, and energy harvesting. Kindly refer to the publication page to review the research endeavors in this field.

Shape memory alloys possess various crystalline states which can be accessed by heating and cooling, making them attractive due to their beneficial properties such as high strength-to-weight ratio, durability, and compact structure. Nevertheless, the limited contraction of shape memory alloy wires can be overcome by incorporating a passive base layer in a morphing configuration to enhance the deformation. To enable self-sensing of the actuator's configuration, voltage probes can be included at the base of a unimorph shape memory alloy actuator. Furthermore, machine learning techniques can be employed to correct significant errors and predict the location of the actuator's end. These actuators can also be viewed through the prism of reservoir computing by designing them to perform both computation and actuation simultaneously. Kindly refer to the publication page to review the research endeavors in this field.

Lamb waves, a type of guided wave that propagates in solid materials, have garnered significant interest for their potential applications in non-destructive testing, structural health monitoring, energy harvesting, and sensing. However, their effectiveness is often limited by factors such as attenuation and dispersion. In this research, acoustic metamaterials are designed as a solution to overcome these limitations and enhance the sensing capabilities with Lamb waves. Acoustic metamaterials are engineered materials with unique properties that can manipulate the propagation of sound waves in desired ways, offering unprecedented control over wave behavior. The goal in this direction is to use vibration techniques, nonlinearity, and data driven optimization method to design acoustic metamaterials with increased sensing abilities. These methods would be applicable to vibrational energy harvesting as well. The manuscripts in this direction are in preparation.