More Like this

1. Predicting Materials Parameters in Colloidal Quantum Dot Photovoltaic Devices Using Machine Learning Models Trained On Experimental Data

   https://doi.org/10.1109/PVSC48317.2022.9938867  

   Lee, Hoon Jeong; Hofelmann, Ariana B.; Lin, Yida; Thon, Susanna M. (June 2022, 2022 IEEE 49th Photovoltaics Specialists Conference (PVSC))

   Numerous characterization techniques have been developed over the last century, which have advanced progress on the development of a variety of photovoltaic technologies. However, this multitude of techniques leads to increasing experimental costs and complexity. It would be useful to have an approach that does not require the time commitment or operation costs to directly learn and implement every new measurement technique. Herein, we explore several machine learning (ML) models that output complex materials parameters, such as electronic trap state density, solely using illuminated current-voltage curves. This greatly reduces both the complexity and cost of the characterization process. Current-voltage curves were chosen as the only input to our models because this type of measurement is relatively simple to perform and most photovoltaic research labs already collect this information on all devices. We compare several different ML network architectures, all of which are trained on experimental data from PbS colloidal quantum dot thin film solar cells. We predict values for underlying materials parameters and compare them to experimentally measured results. 

   more » « less
2. An information-theoretic approach to study spatial dependencies in small datasets

   https://doi.org/10.1098/rspa.2020.0113  

   Porfiri, Maurizio; Ruiz Marín, Manuel (October 2020, Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences)

   null (Ed.)

   From epidemiology to economics, there is a fundamental need of statistically principled approaches to unveil spatial patterns and identify their underpinning mechanisms. Grounded in network and information theory, we establish a non-parametric scheme to study spatial associations from limited measurements of a spatial process. Through the lens of network theory, we relate spatial patterning in the dataset to the topology of a network on which the process unfolds. From the available observations of the spatial process and a candidate network topology, we compute a mutual information statistic that measures the extent to which the measurement at a node is explained by observations at neighbouring nodes. For a class of networks and linear autoregressive processes, we establish closed-form expressions for the mutual information statistic in terms of network topological features. We demonstrate the feasibility of the approach on synthetic datasets comprising 25–100 measurements, generated by linear or nonlinear autoregressive processes. Upon validation on synthetic processes, we examine datasets of human migration under climate change in Bangladesh and motor vehicle deaths in the United States of America. For both these real datasets, our approach is successful in identifying meaningful spatial patterns, begetting statistically-principled insight into the mechanisms of important socioeconomic problems. 

   more » « less
3. Learned load balancing

   https://doi.org/10.1016/J.TCS.2024.114611  

   Chang, Brian; Subramanian, Kausik; D'Antoni, Loris; Akella, Aditya (July 2024, Theoretical Computer Science)

   Effectively balancing traffic in datacenter networks is a crucial operational goal. Most existing load balancing approaches are handcrafted to the structure of the network and/or network workloads. Thus, new load balancing strategies are required if the underlying network conditions change, e.g., due to hard or grey failures, network topology evolution, or workload shifts. While we can theoretically derive the optimal load balancing strategy by solving an optimization problem given certain traffic and topology conditions, these problems take too much time to solve and makes the derived solution stale to deploy. In this paper, we describe a load balancing scheme Learned Load Balancing (LLB), which is a general approach to finding an optimal load balancing strategy for a given network topology and workload, and is fast enough in practice to deploy the inferred strategies. LLB uses deep supervised learning techniques to learn how to handle different traffic patterns and topology changes, and adapts to any failures in the underlying network. LLB leverages emerging trends in network telemetry, programmable switching, and “smart” NICs. Our experiments show that LLB performs well under failures and can be expanded to more complex, multi-layered network topologies. We also prototype neural network inference on smartNICs to demonstrate the workability of LLB. 

   more » « less
4. An MILP Approach for Distribution Grid Topology Identification using Inverter Probing

   https://doi.org/10.1109/PTC.2019.8810900  

   Taheri, Sina; Kekatos, Vassilis; Cavraro, Guido (June 2019, IEEE Power Tech Conference)

   Although knowing the feeder topology and line impedances is a prerequisite for solving any grid optimization task, utilities oftentimes have limited or outdated information on their electric network assets. Given the rampant integration of smart inverters, we have previously advocated perturbing their power injections to unveil the underlying grid topology using the induced voltage responses. Under an approximate grid model, the perturbed power injections and the collected voltage deviations obey a linear regression setup, where the unknown is the vector of line resistances. Building on this model, topology processing can be performed in two steps. Given a candidate radial topology, the line resistances can be estimated via a least-squares (LS) fit on the probing data. The topology attaining the best fit can be then selected. To avoid evaluating the exponentially many candidate topologies, this two-step approach is uniquely formulated as a mixed-integer linear program (MILP) using the McCormick relaxation. If the recovered topology is not radial, a second, computationally more demanding MILP confines the search only within radial topologies. Numerical tests explain how topology recovery depends on the noise level and probing duration, and demonstrate that the first simpler MILP yields a tree topology in 90% of the cases tested. 

   more » « less
5. Edge-Based Heuristics for Optimizing Shortcut-Augmented Topologies for HPC Interconnects

   https://doi.org/10.3390/electronics11172778  

   Fuad, Kazi Ahmed; Zeng, Kai; Chen, Lizhong (September 2022, Electronics)

   Interconnection network topology is critical for the overall performance of HPC systems. While many regular and irregular topologies have been proposed in the past, recent work has shown the promise of shortcut-augmented topologies that offer multi-fold reduction in network diameter and hop count over conventional topologies. However, the large number of possible shortcuts creates an enormous design space for this new type of topology, and existing approaches are extremely slow and do not find shortcuts that are globally optimal. In this paper, we propose an efficient heuristic approach, called EdgeCut, which generates high-quality shortcut-augmented topologies. EdgeCut can identify more globally useful shortcuts by making its considerations from the perspective of edges instead of vertices. An additional implementation is proposed that approximates the costly all-pair shortest paths calculation, thereby further speeding up the scheme. Quantitative comparisons over prior work show that the proposed approach achieves a 1982× reduction in search time while generating better or equivalent topologies in 94.9% of the evaluated cases. 

   more » « less