arXiv subjects

Michael Cooper

Publications and source records attributed to Michael Cooper.

At least 19 recordsLinked to original sources

arXiv2026

Fast, low noise, megapixel detector and readout systems for future X-ray astronomy missions

Next-generation strategic X-ray astronomy missions will require the simultaneous achievement of high angular resolution, large effective collecting area, and wide-field imaging with large-format focal plane detectors. Realizing the associated science objectives--ranging from precision measurements of bright point sources to the detection and characterization of faint diffuse emission-places stringent and, in some cases, competing requirements on detector performance. In particular, high frame rates are necessary to mitigate photon pile-up in observations of bright sources and to reduce contamination from particle-induced background in measurements of low surface brightness structures. At the same time, these instruments must preserve excellent soft X-ray response, which places tight constraints on read noise and on the fidelity of event characterization. State-of-the-art X-ray charge-coupled devices (CCDs) approach many of the key performance metrics required for these missions, but readout speed remains a primary limitation. Addressing this gap requires readout architectures that scale to high channel count, sustain high pixel throughput, and preserve the low-noise characteristics needed for soft X-ray sensitivity.

astro-ph.IM↗

arXiv2026

High-speed, low-noise, multi-megapixel CCDs for next generation X-ray observatories

Next generation X-ray observatories require fast, low-noise, low-power, multi-megapixel imaging spectrometers. To meet these demands, the X-ray Astronomy and Observational Cosmology (XOC) Group at Stanford, in partnership with the MIT Kavli Institute and MIT Lincoln Laboratory (MIT-LL), is developing multi-channel X-ray charge-coupled devices (CCDs) and fast readout architectures. We report the energy resolution and noise performance achieved with a full-scale (1440x1440-pixel), 16-channel, front-illuminated MIT-LL CCD detector developed for the Advanced X-ray Imaging Satellite (AXIS) concept, the CCID-100, read out using two Multi-Channel Readout Chip (MCRC) V1 application-specific integrated circuit (ASIC) chips in the new Stanford CCID-100 test setup. We describe an automated method for bias optimization on each CCD channel, and integrated debugging features of the front-end ASIC and readout system. The demonstrated performance confirms that these systems can meet the speed and noise requirements of future strategic X-ray missions.

astro-ph.IM↗

arXiv2026

Output-Stage Design Optimization for High-Sensitivity SiSeRO CCDs and SiSeRO Active Pixel Sensors

The Single electron Sensitive Read Out (SiSeRO) technology is a new device class designed to support the needs of future X-ray and optical astronomical telescopes that will require fast, low-noise, megapixel spectro-imagers. Developed at MIT Lincoln Laboratory, in collaboration with Stanford University and MIT, the first generation SiSeRO-CCD (charge-coupled device) prototypes achieved a charge/current conversion gain of 700$-$800 pA per electron, an equivalent noise charge (ENC) of around 3.5 electrons root mean square (RMS), and a full width half maximum (FWHM) energy resolution of approximately 130 eV at 5.9 keV at a readout speed of 625 kpix/s. Utilizing Repetitive Non-Destructive Readout (RNDR), these same devices also demonstrated sub-electron noise performance (ENC$<$0.5 electrons RMS) at a readout speed of 10 kpix/s. We present the results of device simulations for next-generation SiSeRO CCD output stages that optimize the sensing transistor and its internal gate geometry to enhance noise and speed performance. Further, the goal is to develop a SiSeRO active pixel sensor (APS) that combines the proven X-ray performance of CCDs with the architectural advantages of an APS. Enabling this requires substantial design updates, for example, incorporating two SiSeRO amplifiers side by side on each pixel and shuffling the charge between them to support RNDR. We discuss our device simulation framework and design parameter optimization in the first-generation SiSeRO devices.

astro-ph.IM↗

arXiv2026

RNDR noise modeling in first-generation Single electron Sensitive Readout (SiSeRO) devices

Flagship observatories require single-photon detectors with ultra-fast readout, sub-electron noise performance, and scalable large-format architectures. The X-ray Astronomy and Observational Cosmology group at Stanford, in collaboration with the MIT Kavli Institute and MIT Lincoln Laboratory, is developing readout technologies for next-generation detectors. Prototypes employing Single-electron Sensitive Readout (SiSeRO) amplifiers demonstrate excellent read noise and spectral performance using repetitive non-destructive readout (RNDR), achieving 0.5 e$^-$ noise in under 57 cycles. We have modeled noise for longer RNDR cycles, exploring probabilistic mechanisms such as thermal leakage and impact ionization. Here we present our model results, including statistical limits on dark current-like signals. Maturation of SiSeRO technology will improve detector performance at soft X-ray energies, addressing technology gaps for future X-ray and UV/visible/near-IR observatories.

astro-ph.IM↗

arXiv2026

First results for second generation SiSeRO CCD devices

The Astro2020 Decadal recommended the development of a suite of next generation astronomical observatories spanning the X-ray to near-IR spectrum. These programs require fast, extremely low noise detectors to fulfill their science goals. To address this technology gap, Stanford X-ray Astronomy and Observational Cosmology (XOC) group, MIT Lincoln Laboratory (MIT-LL), and MIT Kavli Institute (MKI) are advancing Single electron Sensitive Read Out (SiSeRO), a multiband detector technology capable of achieving substantially sub-electron noise via Repetitive Non-Destructive Readout (RNDR). We present initial results for our second generation SiSeRO CCDs. We also discuss our test bed, including a readout electronics system capable of accommodating all second-generation SiSeRO CCD variants utilizing the XOC-designed Multi-Channel Readout Chip (MCRC) ASIC.

astro-ph.IM↗

arXiv2026

SurvivalPFN: Amortizing Survival Prediction via In-Context Bayesian Inference

Survival analysis provides a powerful statistical framework for modeling time-to-event outcomes in the presence of censoring. However, selecting an appropriate estimator from the many specialized survival approaches often requires substantial methodological and domain expertise. We introduce SurvivalPFN, a prior-data fitted network that amortizes Bayesian inference for censored observations through in-context learning. SurvivalPFN is pretrained on a diverse family of synthetic, identifiable, and right-censored data-generating processes, enabling it to amortize survival analysis in a single forward pass during inference. As a result, the model adapts to the effective complexity of each dataset without task-specific training or hyperparameter tuning, avoids restrictive parametric assumptions, and produces calibrated survival distributions. In a large-scale benchmark spanning 61 datasets, 21 methods, and 5 evaluation metrics, SurvivalPFN achieves strong predictive performance and often improves upon established survival models. These results suggest that SurvivalPFN offers a principled and practical foundation model for survival analysis, with potential applications in high-impact domains such as healthcare, finance, and engineering (https://github.com/rgklab/SurvivalPFN).

cs.LG↗

arXiv2026

Mitigating Privacy Risk via Forget Set-Free Unlearning

Training machine learning models requires the storage of large datasets, which often contain sensitive or private data. Storing data is associated with a number of potential risks which increase over time, such as database breaches and malicious adversaries. Machine unlearning is the study of methods to efficiently remove the influence of training data subsets from previously-trained models. Existing unlearning methods typically require direct access to the "forget set" -- the data to be forgotten-and organisations must retain this data for unlearning rather than deleting it immediately upon request, increasing risks associated with the forget set. We introduce partially-blind unlearning -- utilizing auxiliary information to unlearn without explicit access to the forget set. We also propose a practical framework Reload, a partially-blind method based on gradient optimization and structured weight sparsification to operationalize partially-blind unlearning. We show that Reload efficiently unlearns, approximating models retrained from scratch, and outperforms several forget set-dependent approaches. On language models, Reload unlearns entities using <0.025% of the retain set and <7% of model weights in <8 minutes on Llama2-7B. In the corrective case, Reload achieves unlearning even when only 10% of corrupted data is identified.

cs.LG↗

arXiv2026

Math Takes Two: A test for emergent mathematical reasoning in communication

Although language models demonstrate remarkable proficiency on mathematical benchmarks, it remains unclear whether this reflects true mathematical reasoning or statistical pattern matching over learning formal syntax. Most existing evaluations rely on symbolic problems grounded in established mathematical conventions, limiting insight into the models' ability to construct abstract concepts from first principles. In this work, we propose Math Takes Two, a new benchmark designed to assess the emergence of mathematical reasoning through communication. Motivated by the hypothesis that mathematical cognition in humans co-evolved with the need for precise communication, our benchmark tests whether two agents, without prior mathematical knowledge, can develop a shared symbolic protocol to solve a visually grounded task where the use of a numerical system facilitates extrapolation. Unlike many current datasets, our benchmark eschews predefined mathematical language, instead requiring agents to discover latent structure and representations from scratch. Math Takes Two thus provides a novel lens through which to develop and evaluate models with emergent numerical reasoning capabilities.

cs.AI↗

arXiv2025

The high-speed X-ray camera on AXIS: design and performance updates

AXIS, a Probe mission concept now in a Phase A study, will provide transformative studies of high-energy astrophysical phenomena thanks to its high-resolution X-ray spectral imaging. These capabilities are enabled by improvements to the mirror design that greatly increase the X-ray throughput per unit mass; and to the detector system, which operates more than an order of magnitude faster than heritage instruments while maintaining excellent spectral performance. We present updates to the design of the AXIS High-Speed Camera, a collaborative effort by MIT, Stanford University, the Pennsylvania State University, and the Southwest Research Institute. The camera employs large-format MIT Lincoln Laboratory CCDs that feature multiple high-speed, low-noise output amplifiers and an advanced single-layer polysilicon gate structure for fast, low-power clock transfers. A first lot of prototype CCID100 CCDs has completed fabrication and will soon begin X-ray performance testing. The CCDs are paired with high-speed, low-noise ASIC readout chips designed by Stanford to provide better performance than conventional discrete solutions at a fraction of the power consumption and footprint. Complementary Front-End Electronics employ state-of-the-art digital video waveform capture and advanced signal processing to further deliver low noise at high speed. The Back-End Electronics provide high-speed identification of candidate X-ray events and transient monitoring that relays fast alerts of changing sources to the community. We highlight updates to our parallel X-ray performance test facilities at MIT and Stanford, and review the current performance of the CCD and ASIC technology from testing of prototype devices. These measurements achieve excellent spectral response at the required readout rate, demonstrating that we will meet mission requirements and enable AXIS to achieve world-class science.

astro-ph.IM↗

arXiv2025

Development and testing of integrated readout electronics for next generation SiSeRO (Single electron Sensitive Read Out) devices

The first generation of Single electron Sensitive Read Out (SiSeRO) amplifiers, employed as on-chip charge detectors for charge-coupled devices (CCDs) have demonstrated excellent noise and spectral performance: a responsivity of around 800 pA per electron, an equivalent noise charge (ENC) of 3.2 electrons root mean square (RMS), and a full width half maximum (FWHM) energy resolution of 130 eV at 5.9 keV for a readout speed of 625 Kpixel/s. Repetitive Non Destructive Readout (RNDR) has also been demonstrated with these devices, achieving an improved ENC performance of 0.36 electrons RMS after 200 RNDR cycles. In order to mature this technology further, Stanford University, in collaboration with MIT Kavli Institute and MIT Lincoln Laboratory, are developing new SiSeRO detectors with improved geometries that should enable greater responsivity and improved noise performance. These include CCD devices employing arrays of SiSeRO amplifiers to optimize high speed, low noise RNDR readout and a proof-of-concept SiSeRO active pixel sensor (APS). To read out these devices, our team has developed a compact, 8-channel, fast, low noise, low power application specific integrated circuit (ASIC) denoted the Multi-Channel Readout Chip (MCRC) that includes an experimental drain current readout mode intended for SiSeRO devices. In this paper, we present results from the first tests of SiSeRO CCD devices operating with MCRC readout, and our designs for next generation SiSeRO devices.

astro-ph.IM↗

arXiv2025

The Curious Language Model: Strategic Test-Time Information Acquisition

Decision-makers often possess insufficient information to render a confident decision. In these cases, the decision-maker can often undertake actions to acquire the necessary information about the problem at hand, e.g., by consulting knowledgeable authorities or by conducting experiments. Importantly, different levers of information acquisition come with different costs, posing the challenge of selecting the actions that are both informative and cost-effective. In this work, we propose CuriosiTree, a heuristic-based, test-time policy for zero-shot information acquisition in large language models (LLMs). CuriosiTree employs a greedy tree search to estimate the expected information gain of each action and strategically chooses actions based on a balance of anticipated information gain and associated cost. Empirical validation in a clinical diagnosis simulation shows that CuriosiTree enables cost-effective integration of heterogenous sources of information, and outperforms baseline action selection strategies in selecting action sequences that enable accurate diagnosis.

cs.LG↗

arXiv2025

Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift

The subpopulationtion shift, characterized by a disparity in subpopulation distributibetween theween the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real-world datasets, covering diverse domains and kinds of subpopulation shift, our method of Diverse Prototypical Ensembles (DPEs) often outperforms the prior state-of-the-art in worst-group accuracy. The code is available at https://github.com/minhto2802/dpe4subpop

cs.LG↗

arXiv2025

Red Teaming Large Language Models for Healthcare

We present the design process and findings of the pre-conference workshop at the Machine Learning for Healthcare Conference (2024) entitled Red Teaming Large Language Models for Healthcare, which took place on August 15, 2024. Conference participants, comprising a mix of computational and clinical expertise, attempted to discover vulnerabilities -- realistic clinical prompts for which a large language model (LLM) outputs a response that could cause clinical harm. Red-teaming with clinicians enables the identification of LLM vulnerabilities that may not be recognised by LLM developers lacking clinical expertise. We report the vulnerabilities found, categorise them, and present the results of a replication study assessing the vulnerabilities across all LLMs provided.

cs.CL↗

arXiv2024

Predicting Long-Term Allograft Survival in Liver Transplant Recipients

Liver allograft failure occurs in approximately 20% of liver transplant recipients within five years post-transplant, leading to mortality or the need for retransplantation. Providing an accurate and interpretable model for individualized risk estimation of graft failure is essential for improving post-transplant care. To this end, we introduce the Model for Allograft Survival (MAS), a simple linear risk score that outperforms other advanced survival models. Using longitudinal patient follow-up data from the United States (U.S.), we develop our models on 82,959 liver transplant recipients and conduct multi-site evaluations on 11 regions. Additionally, by testing on a separate non-U.S. cohort, we explore the out-of-distribution generalization performance of various models without additional fine-tuning, a crucial property for clinical deployment. We find that the most complex models are also the ones most vulnerable to distribution shifts despite achieving the best in-distribution performance. Our findings not only provide a strong risk score for predicting long-term graft failure but also suggest that the routine machine learning pipeline with only in-distribution held-out validation could create harmful consequences for patients at deployment.

cs.LG↗

arXiv2024

Demonstrating sub-electron noise performance in Single electron Sensitive Readout (SiSeRO) devices

Single electron Sensitive Read Out (SiSeRO) is a novel on-chip charge detection technology that can, in principle, provide significantly greater responsivity and improved noise performance than traditional charge coupled device (CCD) readout circuitry. The SiSeRO, developed by MIT Lincoln Laboratory, uses a p-MOSFET transistor with a depleted back-gate region under the transistor channel; as charge is transferred into the back gate region, the transistor current is modulated. With our first generation SiSeRO devices, we previously achieved a responsivity of around 800 pA per electron, an equivalent noise charge (ENC) of 4.5 electrons root mean square (RMS), and a full width at half maximum (FWHM) spectral resolution of 130 eV at 5.9 keV, at a readout speed of 625 Kpixel/s and for a detector temperature of 250 K. Importantly, since the charge signal remains unaffected by the SiSeRO readout process, we have also been able to implement Repetitive Non-Destructive Readout (RNDR), achieving an improved ENC performance. In this paper, we demonstrate sub-electron noise sensitivity with these devices, utilizing an enhanced test setup optimized for RNDR measurements, with excellent temperature control, improved readout circuitry, and advanced digital filtering techniques. We are currently fabricating new SiSeRO detectors with more sensitive and RNDR-optimized amplifier designs, which will help mature the SiSeRO technology in the future and eventually lead to the pathway to develop active pixel sensor (APS) arrays using sensitive SiSeRO amplifiers on each pixel. Active pixel devices with sub-electron sensitivity and fast readout present an exciting option for next generation, large area astronomical X-ray telescopes requiring fast, low-noise megapixel imagers.

astro-ph.IM↗

arXiv2024

Curved detectors for future X-ray astrophysics missions

Future X-ray astrophysics missions will survey large areas of the sky with unparalleled sensitivity, enabled by lightweight, high-resolution optics. These optics inherently produce curved focal surfaces with radii as small as 2 m, requiring a large area detector system that closely conforms to the curved focal surface. We have embarked on a project using a curved charge-coupled device (CCD) detector technology developed at MIT Lincoln Laboratory to provide large-format, curved detectors for such missions, improving performance and simplifying design. We present the current status of this work, which aims to curve back-illuminated, large-format (5 cm x 4 cm) CCDs to 2.5-m radius and confirm X-ray performance. We detail the design of fixtures and the curving process, and present intial results on curving bare silicon samples and monitor devices and characterizing the surface geometric accuracy. The tests meet our accuracy requirement of <5 $\mu$m RMS surface non-conformance for samples of similar thickness to the functional detectors. We finally show X-ray performance measurements of planar CCDs that will serve as a baseline to evaluate the curved detectors. The detectors exhibit low noise, good charge-transfer efficiency, and excellent, uniform spectroscopic performance, including in the important soft X-ray band.

astro-ph.IM↗

arXiv2024

InterpreTabNet: Distilling Predictive Signals from Tabular Data by Salient Feature Interpretation

Tabular data are omnipresent in various sectors of industries. Neural networks for tabular data such as TabNet have been proposed to make predictions while leveraging the attention mechanism for interpretability. However, the inferred attention masks are often dense, making it challenging to come up with rationales about the predictive signal. To remedy this, we propose InterpreTabNet, a variant of the TabNet model that models the attention mechanism as a latent variable sampled from a Gumbel-Softmax distribution. This enables us to regularize the model to learn distinct concepts in the attention masks via a KL Divergence regularizer. It prevents overlapping feature selection by promoting sparsity which maximizes the model's efficacy and improves interpretability to determine the important features when predicting the outcome. To assist in the interpretation of feature interdependencies from our model, we employ a large language model (GPT-4) and use prompt engineering to map from the learned feature mask onto natural language text describing the learned signal. Through comprehensive experiments on real-world datasets, we demonstrate that InterpreTabNet outperforms previous methods for interpreting tabular data while attaining competitive accuracy.

cs.LG↗

arXiv2023

Copula-Based Deep Survival Models for Dependent Censoring

A survival dataset describes a set of instances (e.g. patients) and provides, for each, either the time until an event (e.g. death), or the censoring time (e.g. when lost to follow-up - which is a lower bound on the time until the event). We consider the challenge of survival prediction: learning, from such data, a predictive model that can produce an individual survival distribution for a novel instance. Many contemporary methods of survival prediction implicitly assume that the event and censoring distributions are independent conditional on the instance's covariates - a strong assumption that is difficult to verify (as we observe only one outcome for each instance) and which can induce significant bias when it does not hold. This paper presents a parametric model of survival that extends modern non-linear survival analysis by relaxing the assumption of conditional independence. On synthetic and semi-synthetic data, our approach significantly improves estimates of survival distributions compared to the standard that assumes conditional independence in the data.

cs.LG↗