In this article, we report the growth of Cd0.9Zn0.1Te0.97Se0.03 (CZTS) wide bandgap semiconductor single crystals for room temperature gamma-ray detection using a modified vertical Bridgman method. Charge transport properties measured in the radiation detectors, fabricated from the grown CZTS crystals, indicated signs of hole trapping. Hole traps inhibit high-resolution radiation detection especially for energetic gamma rays. Machine learning (ML) applications are gaining tremendous impetus in improving device and sensor performance by compensating for limitations arising from such intrinsic material properties. In this article, we describe a deep convolutional neural network (CNN) that has demonstrated remarkable efficiency in identifying the energy of a gamma photon detected by a CZTS detector. The CNN has been trained using simulated data that resemble output pulses from actual CZTS detectors when exposed to 662-keV gamma photons. The device properties required for the simulation have been derived from radiation detection measurements on a real Cd0.9Zn0.1Te0.97Se0.03 detector fabricated in our laboratory. The CNN has been trained with detector pulses arising through photoelectric (PE) and Compton scattering (CS) separately. The percentage error in predicting the detected energies, within an extremely small duration of 0.28 ms, was found to be lower than 0.1{\%} for gamma energies above 50 keV and for training datasets containing PE and CS events separately. The CNN was also validated for a mixed PE and CS dataset to obtain a prediction error of 1{\%}. The effect of detector resolution on the efficiency of the CNN was also explored.
Publications
2022
Circadian rhythms are fundamental to biology and medicine and today these can be studied at the molecular level in high-throughput fashion using various omic technologies. We briefly present two resources for the study of circadian omic (e.g. transcriptomic, metabolomic, proteomic) time series. First, BIO\_CYCLE is a deep-learning-based program and web server that can analyze omic time series and statistically assess their periodic nature and, when periodic, accurately infer the corresponding period, amplitude, and phase. Second, CircadiOmics is the larges annotated repository of circadian omic time series, containing over 260 experiments and 90 million individual measurements, across multiple organs and tissues, and across 9 different species. In combination, these tools enable powerful bioinformatics and systems biology analyses. The are currently being deployed in a host of different projects where they are enabling significant discoveries: both tools are publicly available over the web at: http://circadiomics.ics.uci.edu/.
Modern artificial intelligence (AI) methods have been used to solve problems that many humans struggle to solve. This opens up new opportunities for knowledge discovery and education. We demonstrate ALLURE, a collaborative educational AI system for learning to solve the Rubik’s cube that is designed to help students improve their problem solving skills. ALLURE can both find its own strategies for solving the Rubik’s cube and explain those strategies to humans. In the future, ALLURE will also be able to collaborate with humans by building on user-provided strategies for solving the Rubik’s cube and as well as generalize to other search and automated planning problems. Interaction between AI and user is facilitated by visual and natural language modalities.
Rubik’s Cube (RC) is a popular puzzle that is also computationally hard to solve. In this demonstration, we introduce the first PDDL formulation for the 3-dimension RC and solve it with off-the-shelf Fast-Downward planner. We also create a plan executor and visualizer to show how the plan achieves the intended goal. Our system has two audiences:(a) planning researchers who can explore a hard problem, and (b) RC learners wanting to learn how to solve the puzzle at their own pace.
The hippocampus is critical to the temporal organization of our experiences. Although this fundamental capacity is conserved across modalities and species, its underlying neuronal mechanisms remain unclear. Here we recorded hippocampal activity as rats remembered an extended sequence of nonspatial events unfolding over several seconds, as in daily life episodes in humans. We then developed statistical machine learning methods to analyze the ensemble activity and discovered forms of sequential organization and coding important for order memory judgments. Specifically, we found that hippocampal ensembles provide significant temporal coding throughout nonspatial event sequences, differentiate distinct types of task-critical information sequentially within events, and exhibit theta-associated reactivation of the sequential relationships among events. We also demonstrate that nonspatial event representations are sequentially organized within individual theta cycles and precess across successive cycles. These findings suggest a fundamental function of the hippocampal network is to encode, preserve, and predict the sequential order of experiences.
2021
Deep reinforcement learning has been shown to be able to train deep neural networks to implement effective heuristic functions that can be used with A* search to solve problems with large state spaces. However, these learned heuristic functions are not guaranteed to be admissible. We introduce approximately admissible conversion, an algorithm that can convert any inadmissible heuristic function into a heuristic function that is admissible in the vast majority of cases with no domain-specific heuristic information. We apply approximately admissible conversion to heuristic functions parameterized by deep neural networks and show that these heuristic functions can be used to find optimal solutions, or bounded suboptimal solutions, even when doing a batched version of A* search. We test our method on the 15-puzzle and 24-puzzle and obtain a heuristic function that is empirically admissible over 99.99% of the time and that finds optimal solutions for 100% of all test configurations. To the best of our knowledge, this is the first demonstration that approximately admissible heuristics can be obtained using deep neural networks in a domain independent fashion.
