Search for: All records

Recurrent Neural Networks (RNNs) are important tools for processing sequential data such as time-series or video. Interpretability is defined as the ability to be understood by a person and is different from explainability, which is the ability to be explained in a mathematical formulation. A key interpretability issue with RNNs is that it is not clear how each hidden state per time step contributes to the decision-making process in a quantitative manner. We propose NeuroView-RNN as a family of new RNN architectures that explains how all the time steps are used for the decision-making process. Each member of the family is derived from a standard RNN architecture by concatenation of the hidden steps into a global linear classifier. The global linear classifier has all the hidden states as the input, so the weights of the classifier have a linear mapping to the hidden states. Hence, from the weights, NeuroView-RNN can quantify how important each time step is to a particular decision. As a bonus, NeuroView-RNN also offers higher accuracy in many cases compared to the RNNs and their variants. We showcase the benefits of NeuroView-RNN by evaluating on a multitude of diverse time-series datasets. 

Deep neural networks have become essential for numerous applications due to their strong empirical performance such as vision, RL, and classification. Unfortunately, these networks are quite difficult to interpret, and this limits their applicability in settings where interpretability is important for safety, such as medical imaging. One type of deep neural network is neural tangent kernel that is similar to a kernel machine that provides some aspect of interpretability. To further contribute interpretability with respect to classification and the layers, we develop a new network as a combination of multiple neural tangent kernels, one to model each layer of the deep neural network individually as opposed to past work which attempts to represent the entire network via a single neural tangent kernel. We demonstrate the interpretability of this model on two datasets, showing that the multiple kernels model elucidates the interplay between the layers and predictions. 

Solid-state thermionic devices based on van der Waals structures were proposed for nanoscale thermal to electrical energy conversion and integrated electronic cooling applications. We study thermionic cooling across gold-graphene-WSe 2 -graphene-gold structures computationally and experimentally. Graphene and WSe 2 layers were stacked, followed by deposition of gold contacts. The I - V curve of the structure suggests near-ohmic contact. A hybrid technique that combines thermoreflectance and cooling curve measurements is used to extract the device ZT . The measured Seebeck coefficient, thermal and electrical conductance, and ZT values at room temperatures are in agreement with the theoretical predictions using first-principles calculations combined with real-space Green’s function formalism. This work lays the foundation for development of efficient thermionic devices. 

Hybrid organic-inorganic materials are among the latest class of materials proposed for thermoelectric applications. The organic-inorganic interface is critical in determining the effective transport properties of the hybrid material. We study the thermoelectric properties of the tetrafluoro-tetracyanoquinodimethane (F4TCNQ)–silicon interface. Transfer of electrons from silicon to F4TCNQ results in holes trapped within the screening length of the interface that can move parallel to the interface. We measure the response of these trapped charges to applied temperature differential and compare the thermoelectric transport properties of the silicon with and without F4TCNQ. The results confirm the presence of interface charges and demonstrate an enhanced interface thermoelectric power factor. These outcomes of this study could be used in designing 3D hybrid structures with closely packed interfaces to replicate a bulk thermoelectric material. 

Abstract A β‐FeSi₂–SiGe nanocomposite is synthesized via a react/transform spark plasma sintering technique, in which eutectoid phase transformation, Ge alloying, selective doping, and sintering are completed in a single process, resulting in a greatly reduced process time and thermal budget. Hierarchical structuring of the SiGe secondary phase to achieve coexistence of a percolated network with isolated nanoscale inclusions effectively decouples the thermal and electrical transport. Combined with selective doping that reduces conduction band offsets, the percolation strategy produces overall electron mobilities 30 times higher than those of similar materials produced using typical powder‐processing routes. As a result, a maximum thermoelectric figure of meritZTof ≈0.7 at 700 °C is achieved in the β‐FeSi₂–SiGe nanocomposite.