Recent publications
216
Background: Colorectal cancer (CRC) is the second leading cause of cancer-related deaths. We previously found that delayed diagnosis due to lack of radiological identification results in significantly worse outcome for patients. We had developed a rudimentary, AI second observer which demonstrated potential for detecting CRC on routine CT abdomen/pelvis (CTAP). However, the AI algorithm detected many false positives. In this study, we analyzed the data using TCIA as test cases and evaluated whether patient peritoneal fat content influenced the false positive rate. This could serve as a guide for future training of AI second observer to minimize false positive detection. Methods: 2D U-Net convolutional neural network (CNN) containing 31 million trainable parameters was trained with 58 CRC CT images from Banner MD Anderson (AZ) and MD Anderson Cancer Center (TX) (51 used for training and 7 for validation) and 59 normal CT scans from Banner MD Anderson Cancer Center. 18 of the 25 CRC cases from public domain data (The Cancer Genome Atlas) were used to evaluate the performance of the models (5 had no identifiable cancer and 2 were rejected for having no contrast). The CRC was segmented using ITK-SNAP open-source software (v. 3.8). To apply the deep ensemble approach, five CNN models were trained independently with random initialization using the same U-Net architect and the same training data. Given a testing CT scan, each of the five trained CNN models was applied to produce tumor segmentation for the testing CT scan. The tumor segmentation results produced by the trained CNN models were then fused using a simple majority voting rule (up to 2 voters) to produce consensus tumor segmentation results. The segmentation was analyzed for the number and location of false positives per case. The peritoneal fat content was classified at the level of aortic bifurcation by the distance of fat between adjacent small bowel loops (≤ or > 1 cm). Chi-square test was performed testing fat volume and number of voters as the intervention. Results: Our results showed that the higher volume of peritoneal fat (> 1 cm, N=6) decreases the rate of false positive compared with low volume (≤ 1 cm, N=12, p=0.013). When comparing between having one voter and two voter ensemble using low fat volume data, two voter ensemble also decreased the number of false positives but not statistically significant (p=0.286). Conclusions: Our results show that AI-based second observer generates more false positives when patients have lower peritoneal fat volume; this implies that future training may require higher percentage of cases with low peritoneal fat to improve second observer precision. Our analysis also showed that increasing the number of voter in the ensemble also decrease the number of false positives per case.
The development of high-brightness electron sources is critical to state-of-the-art electron accelerator applications like X-ray free electron laser (XFEL) and ultra-fast electron microscopy. Cesium telluride is chosen as the electron source material for multiple cutting-edge XFEL facilities worldwide. This manuscript presents the first demonstration of the growth of highly crystalized and epitaxial cesium telluride thin films on 4H-SiC and graphene/4H-SiC substrates with ultrasmooth film surfaces. The ordering of the film was characterized by in situ reflection high energy electron diffraction and multiple X-ray diagnostics. The results of the quantum efficiency performance for epitaxial cesium telluride photocathodes are also reported.
Long‐term eddy covariance (EC) data are crucial for understanding the impact of global change on ecosystem functions. However, EC data often contain long gaps, particularly in tropical dry forests (TDF) due to seasonality and El Niño‐Southern Oscillation (ENSO) phases. These factors create high variability, complex dependencies, and dynamic flux footprints. No current gap‐filling method adequately addresses long gaps in TDFs. This study introduces a novel framework for addressing this issue by (a) defining gap sizes by their relative percentages, (b) training, tuning, and evaluating two machine learning (ML) models: MissForest for short gaps and Prophet for intermediate and long gaps, and (c) predicting half‐hourly EC data from 2013 to 2022 for six EC variables, where actual gap data sets ranged from 26.6% to 28.4%, at TDF in Costa Rica. Results indicate that MissForest excelled at filling short gaps (≤5%, R² = 0.76 and Nash‐Sutcliffe efficiency (NSE) = 0.71), while Prophet performed exceptionally well for gaps between 5% and 10% (R² = 0.72 and NSE = 0.67). However, both models struggled with gaps between 10% and 13%. Validation showed R² values of 0.79, 0.88, and 0.77 for CO₂ flux, sensible heat flux, and latent heat flux, respectively, with corresponding NSE values of 0.78, 0.86, and 0.72, and normalized root mean squared error (NRMSE) around 2E‐4. Additionally, to validate our results, we applied our approach at three EC sites with different ecological conditions, demonstrating robust performance. This study presents a reliable ML approach for imputing long gaps in EC data, which can be applied to sites with strong variability.
Computing nonlinear functions over multilinear forms is a general problem with applications in risk analysis. For instance in the domain of energy economics, accurate and timely risk management demands for efficient simulation of millions of scenarios, largely benefiting from computational speedups. We develop a novel hybrid quantum–classical algorithm based on polynomial approximation of nonlinear functions, computed through Quantum Hadamard Products, and we rigorously assess the conditions for its end-to-end speedup for different implementation variants against classical algorithms. In our setting, a quadratic quantum speedup, up to polylogarithmic factors, can be proven only when forms are bilinear and approximating polynomials have second degree, if efficient loading unitaries are available for the input data sets. We also enhance the bidirectional encoding, that allows tuning the balance between circuit depth and width, proposing an improved version that can be exploited for the calculation of inner products. Lastly, we exploit the dynamic circuit capabilities, recently introduced on IBM Quantum devices, to reduce the average depth of the Quantum Hadamard Product circuit. A proof of principle is implemented and validated on IBM Quantum systems.
Electron transport measurements on 60-nm-thick multilayers containing N = 2–58 individual Ru and Co layers are employed to quantify the specific resistance of Ru/Co interfaces. Sputter deposition on Al2O3(0001) at Ts = 400 °C leads to a 0001 preferred orientation with x-ray diffraction (XRD) Ru and Co 0002 peaks that shift closer to each other with increasing N, suggesting interfacial intermixing. The intermixing is quantified by x-ray reflectivity (XRR) and confirmed by an XRD Ru/Co alloy peak that develops during in situ synchrotron annealing as well as for deposition at a higher Ts = 600 °C. The room-temperature resistivity increases from 15.0 to 47.5 μΩ cm with decreasing superlattice period Λ = 60–2 nm. This is attributed to increasing electron scattering at the intermixed metal interfaces. The transport data are well described by a parallel conductor model that treats metal layers and the intermixed alloy as parallel resistors, where the resistivity of the intermixed alloy of 60.4 μΩ cm is determined from a co-deposited Ru/Co sample. Data fitting provides values for the effective thickness of the intermixed interface of 16.8 nm, in good agreement with the XRR value, yielding a Ru/Co contact resistance of 8.5 × 10⁻¹⁵ Ω m² for interfaces deposited at 400 °C. The overall results show that the Ru/Co contact resistance is dominated by a high-resistivity interfacial alloy and, therefore, is a strong function of the deposition process, particularly the processing temperature.
Quantum neuromorphic computing (QNC) is a sub-field of quantum machine learning (QML) that capitalizes on inherent system dynamics. As a result, QNC can run on contemporary, noisy quantum hardware and is poised to realize challenging algorithms in the near term. One key issue in QNC is the characterization of the requisite dynamics for ensuring expressive quantum neuromorphic computation. We address this issue by adapting previous proposals of quantum perceptrons (QPs), a quantum version of a simplistic model for neural computation, to the QNC setting. Our QPs compute based on the analog dynamics of interacting qubits with tunable coupling constants. We show that QPs are, with restricted resources, a quantum equivalent to the classical perceptron, a simple mathematical model for a neuron that is the building block of various machine learning architectures. Moreover, we show that QPs are theoretically capable of producing any unitary operation. Thus, QPs are computationally more expressive than their classical counterparts. As a result, QNC architectures built using our QPs are, theoretically, universal. We introduce a technique for mitigating barren plateaus in QPs called entanglement thinning. We demonstrate QPs’ effectiveness by applying them to numerous QML problems, including calculating the inner products between quantum states, energy measurements, and time reversal. Finally, we discuss potential implementations of QPs and how they can be used to build more complex QNC architectures such as quantum reservoir computers.
Large language models (LLMs), with their remarkable generative capacities, have greatly impacted a range of fields, but they face scalability challenges due to their large parameter counts, which result in high costs for training and inference. The trend of increasing model sizes is exacerbating these challenges, particularly in terms of memory footprint, latency and energy consumption. Here we explore the deployment of ‘mixture of experts’ (MoEs) networks—networks that use conditional computing to keep computational demands low despite having many parameters—on three-dimensional (3D) non-volatile memory (NVM)-based analog in-memory computing (AIMC) hardware. When combined with the MoE architecture, this hardware, utilizing stacked NVM devices arranged in a crossbar array, offers a solution to the parameter-fetching bottleneck typical in traditional models deployed on conventional von-Neumann-based architectures. By simulating the deployment of MoEs on an abstract 3D AIMC system, we demonstrate that, due to their conditional compute mechanism, MoEs are inherently better suited to this hardware than conventional, dense model architectures. Our findings suggest that MoEs, in conjunction with emerging 3D NVM-based AIMC, can substantially reduce the inference costs of state-of-the-art LLMs, making them more accessible and energy-efficient.
Artificial intelligence (AI) has transformed infectious disease control, enhancing rapid diagnosis and antibiotic discovery. While conventional tests delay diagnosis, AI-driven methods like machine learning and deep learning assist in pathogen detection, resistance prediction, and drug discovery. These tools improve antibiotic stewardship and identify effective compounds such as antimicrobial peptides and small molecules. This review explores AI applications in diagnostics, therapy, and drug discovery, emphasizing both strengths and areas needing improvement.
Problem definition: In this paper, we present a reinforcement learning (RL)-based framework for optimizing long-term discounted reward problems with large combinatorial action space and state dependent constraints. These characteristics are common to many operations management problems, for example, network inventory replenishment, where managers have to deal with uncertain demand, lost sales, and capacity constraints that results in more complex feasible action spaces. Our proposed programmable actor RL (PARL) uses a deep-policy iteration method that leverages neural networks to approximate the value function and combines it with mathematical programming and sample average approximation to solve the per-step-action optimally while accounting for combinatorial action spaces and state-dependent constraint sets. Methodology/results: We then show how the proposed methodology can be applied to complex inventory replenishment problems where analytical solutions are intractable. We also benchmark the proposed algorithm against state-of-the-art RL algorithms and commonly used replenishment heuristics and find that the proposed algorithm considerably outperforms existing methods by as much as 14.7% on average in various complex supply chain settings. Managerial implications: We find that this improvement in performance of PARL over benchmark algorithms can be directly attributed to better inventory cost management, especially in inventory constrained settings. Furthermore, in the simpler setting where optimal replenishment policy is tractable or known near optimal heuristics exist, we find that the RL-based policies can learn near optimal policies. Finally, to make RL algorithms more accessible for inventory management researchers, we also discuss the development of a modular Python library that can be used to test the performance of RL algorithms with various supply chain structures. This library can spur future research in developing practical and near-optimal algorithms for inventory management problems.
Supplemental Material: The online appendix is available at https://doi.org/10.1287/msom.2022.0617 .
Although optimal control (OC) has been studied in stochastic thermodynamics for systems with continuous state variables, less is known in systems with discrete state variables, such as chemical reaction networks (CRNs). Here, we develop a general theoretical framework to study OC of CRNs for changing the system from an initial distribution of states to a final distribution with minimum dissipation. We derive a "Kirchhoff's law" for the probability current in the adiabatic limit, from which the optimal kinetic rates are determined analytically for any given probability trajectory allowed by local rate constraints. By using the optimal rates, we show that the total dissipation is determined by a L_{2}-distance measure in the probability space and derive an analytical expression for the metric tensor that depends on the probability distribution, network topology, and capacity of each link. Minimizing the total dissipation leads to the geodesic trajectory in the probability space and the corresponding OC protocol is determined by the Kirchhoff's law. To demonstrate our general approach, we use it to find a lower bound for the minimum dissipation that is tighter than existing bounds obtained with only global constraints in the adiabatic limit. We also apply it to simple networks, e.g., fully connected three-state CRNs with different local constraints and show that indirect pathway and nonfunctional transient state can play a crucial role in switching between different probability distributions efficiently. Future directions in studying OC in CRNs by using our general framework are discussed.
The electrical resistivity of conventional metals such as copper is known to increase in thin films as a result of electron-surface scattering, thus limiting the performance of metals in nanoscale electronics. Here, we find an unusual reduction of resistivity with decreasing film thickness in niobium phosphide (NbP) semimetal deposited at relatively low temperatures of 400°C. In films thinner than 5 nanometers, the room temperature resistivity (~34 microhm centimeters for 1.5-nanometer-thick NbP) is up to six times lower than the resistivity of our bulk NbP films, and lower than conventional metals at similar thickness (typically about 100 microhm centimeters). The NbP films are not crystalline but display local nanocrystalline, short-range order within an amorphous matrix. Our analysis suggests that the lower effective resistivity is caused by conduction through surface channels, together with high surface carrier density and sufficiently good mobility as the film thickness is reduced. These results and the fundamental insights obtained here could enable ultrathin, low-resistivity wires for nanoelectronics beyond the limitations of conventional metals.
A modular 4.26Mb SRAM based on a 82Kb/block structure with mixed signal logic is fabricated, characterized, and demonstrated with full functionality in a 3nm nanosheet (NS) technology. Designed macros utilize new circuits for supply boosting, read, and write assist techniques. The proposed circuits are evaluated extensively and compared to prior techniques. Statistical simulations are used to predict the benefits of these circuits in the context of dual supply use. Through programmable local clock and wordline (WL) pulsewidths, SRAM cell margins and speeds are demonstrated through hardware measurement. Stability assists as well as dual supply techniques are used to demonstrate how noise can be suppressed during traditional memory operations (single WL on), as well as to support mixed-signal logic block operation (multiple WLs on). Functionality is shown down to a cell supply of 0.45V with an estimated margin/speed of 6 GHz for SRAM cells (High Density-0.026μm2, and High Current -0.032μ m2).
Alkali antimonides are well established as high efficiency, low intrinsic emittance photocathodes for accelerators and photon detectors. However, conventionally grown alkali antimonide films are polycrystalline with surface disorder and roughness that can limit achievable beam brightness. Ordering the crystalline structure of alkali antimonides has the potential to deliver higher brightness electron beams by reducing surface disorder and enabling the engineering of material properties at the level of atomic layers. In this report, we demonstrate the growth of ordered Cs3Sb films on single crystal substrates 3C-SiC and graphene-coated 4H-SiC using pulsed laser deposition and conventional thermal evaporation growth techniques. The crystalline structures of the Cs3Sb films were examined using reflection high energy electron diffraction and x-ray diffraction diagnostics, while film thickness and roughness estimates were made using x-ray reflectivity. With these tools, we observed ordered domains in less than 10 nm thick films with quantum efficiencies greater than 1% at 530 nm. Moreover, we identify structural features such as Laue oscillations indicative of highly ordered films. We found that Cs3Sb films grew with flat, fiber-textured surfaces on 3C-SiC and with multiple ordered domains and sub-nanometer surface roughness on graphene-coated 4H-SiC under our growth conditions. We identify the crystallographic orientations of Cs3Sb grown on graphene-coated 4H-SiC substrates and discuss the significance of examining the crystal structure of these films for growing epitaxial heterostructures in future experiments.
We present a D -band power amplifier (PA), implemented in Teledyne (TSC)-250-nm indium phosphide (InP) technology, that produces record 27.2-dBm output power and 14.9% associated power-added efficiency (PAE) at 150 GHz. The measured saturated output power and PAE exceed 26 dBm and 12.5% over 126–150 GHz. The output stage power combines sixteen common-base (CB) cells, each having a total emitter length of 24 μm. Each power cell is independently biased by an adaptive bias network (ABN) that prevents thermal runaway and increases bias currents with increased RF power to delay the gain compression. The PA occupies 3 mm2 of the die area. To the authors’ knowledge, this work achieves the highest output power among D -band PAs implemented in any technology.
A 21–27-GHz frequency quadrupler in the 0.13-μm SiGe BiCMOS technology with the 0-dBm output power ( P OUT ) and 40-dBc harmonic rejection ratio ( HRR ) is presented. A method for load—pull-based output network design is introduced to co-optimize HRR and P OUT ; as a result, the design achieves flat and high HRR and P OUT across 25% bandwidth and a wide input power ( P IN ) range. This article also discusses the quadrupler’s P OUT and HRR specifications in the context of its integration within a phased-array antenna module (PAAM). We designed two versions of the 64-element wideband 5G phased-array PAAM, one including and one excluding the quadrupler, to demonstrate the minimal impact of the quadrupler on the output spectrum. We also measure the spur performance in dual-polarization mode to evaluate cross-polarization spurs. The spurious emissions across P OUT range of the phased array is better than -20 dBm/MHz, well below the 3GPP 5G FR2 limit of -15 dBm/MHz. The quadrupler design has the highest HRR performance reported among wideband mmWave quadruplers and thoroughly demonstrates, for the first time, the impact of the local oscillator (LO) frequency multiplier on the performance of a wideband phased-array system.
Digital in-memory compute (IMC) architectures allow for a balance of the high accuracy and precision necessary for many machine learning applications, with high data reuse and parallelism to reduce energy consumption. However, one often overlooked parameter is security, which is necessary to maintain the privacy and integrity of the accelerator. In this work, we propose an IMC macro design that is protected against two types of eavesdropping attacks, passive physical side-channels and memory bus-probing. This is achieved through secure compute that eliminates the need for random bits, local model decryption with a lightweight cipher, and secret key generation reusing existing IMC circuitry. These contributions provide side-channel security against all practical attackers beyond 1 million samples, while still operating without any effect on neural network accuracy at 8.1 TOPS/W energy efficiency.
Institution pages aggregate content on ResearchGate related to an institution. The members listed on this page have self-identified as being affiliated with this institution. Publications listed on this page were identified by our algorithms as relating to this institution. This page was not created or approved by the institution. If you represent an institution and have questions about these pages or wish to report inaccurate content, you can contact us here.
Information
Address
Yorktown Heights, United States