Source author record

Chenhao Zhang

Chenhao Zhang appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

astro-ph.CO Computer Science and Game Theory Computer Vision Machine Learning

Catalog footprint

What is connected

5works

4topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2026arXiv

Barriers to Counterfactual Credit Attribution for Autoregressive Models

Generative AI disrupts the practice of giving credit to work that came before. Ideally, a generative model would give credit to any work on which its output depends in a significant way. \emph{Counterfactual credit attribution} (CCA) is a technical condition formalizing this goal--a relaxation of differential privacy--recently introduced by Livni, Moran, Nissim, and Pabbaraju [2024] who studied it in the PAC learning setting. We initiate the study of CCA generative models. Specifically, we consider autoregressive models giving credit to a deployment-time dataset (e.g., a RAG database). We uncover barriers to two natural approaches to CCA autoregressive models. First, we show that imposing CCA on the underlying next-token predictor does not guarantee that the model is CCA: CCA does not compose autoregressively (unlike DP). Second, we consider a different approach to building CCA models which we call \emph{retrofitting}. Retrofitting takes a model that does not attribute credit, and adds credit onto it. We prove a lower bound for CCA retrofitting under a weak optimality requirement. Given black-box access to the starting model, retrofitting requires query complexity exponential in the length of the model's outputs.

preprint2026arXiv

Instance camera focus prediction for crystal agglomeration classification

Agglomeration refers to the process of crystal clustering due to interparticle forces. Crystal agglomeration analysis from microscopic images is challenging due to the inherent limitations of two-dimensional imaging. Overlapping crystals may appear connected even when located at different depth layers. Because optical microscopes have a shallow depth of field, crystals that are in-focus and out-of-focus in the same image typically reside on different depth layers and do not constitute true agglomeration. To address this, we first quantified camera focus with an instance camera focus prediction network to predict 2 class focus level that aligns better with visual observations than traditional image processing focus measures. Then an instance segmentation model is combined with the predicted focus level for agglomeration classification. Our proposed method has a higher agglomeration classification and segmentation accuracy than the baseline models on ammonium perchlorate crystal and sugar crystal dataset.

preprint2020arXiv

Optimal Budget-Feasible Mechanisms for Additive Valuations

In this paper, we show a tight approximation guarantee for budget-feasible mechanisms with an additive buyer. We propose a new simple randomized mechanism with approximation ratio of $2$, improving the previous best known result of $3$. Our bound is tight with respect to either the optimal offline benchmark, or its fractional relaxation. We also present a simple deterministic mechanism with the tight approximation guarantee of $3$ against the fractional optimum, improving the best known result of $(2+ \sqrt{2})$ for the weaker integral benchmark.

preprint2016arXiv

A Chandra Study of the Image Power Spectra of 41 Cool Core and Non-Cool Core Galaxy Clusters

In this work we propose a new diagnostic to segregate cool core (CC) clusters from non-cool core (NCC) clusters by studying the two-dimensional power spectra of the X-ray images observed with the Chandra X-ray observatory. Our sample contains 41 members ($z=0.01\sim 0.54$), which are selected from the Chandra archive when a high photon count, an adequate angular resolution, a relatively complete detector coverage, and coincident CC-NCC classifications derived with three traditional diagnostics are simultaneously guaranteed. We find that in the log-log space the derived image power spectra can be well represented by a constant model component at large wavenumbers, while at small wavenumbers a power excess beyond the constant component appears in all clusters, with a clear tendency that the excess is stronger in CC clusters. By introducing a new CC diagnostic parameter, i.e., the power excess index (PEI), we classify the clusters in our sample and compare the results with those obtained with three traditional CC diagnostics. We find that the results agree with each other very well. By calculating the PEI values of the simulated clusters, we find that the new diagnostic works well at redshifts up to 0.5 for intermediately sized and massive clusters with a typical Chandra or XMM pointing observation. The new CC diagnostic has several advantages over its counterparts, e.g., it is free of the effects of the commonly seen centroid shift of the X-ray halo caused by merger event, and the corresponding calculation is straightforward, almost irrelevant to the complicated spectral analysis.

preprint2015arXiv

A Chandra Study of Radial Temperature Profiles of the Intra-Cluster Medium in 50 Galaxy Clusters

In order to investigate the spatial distribution of the ICM temperature in galaxy clusters in a quantitative way and probe the physics behind, we analyze the X-ray spectra of a sample of 50 galaxy clusters, which were observed with the Chandra ACIS instrument in the past 15 years, and measure the radial temperature profiles out to $0.45r_{500}$. We construct a physical model that takes into account the effects of gravitational heating, thermal history (such as radiative cooling, AGN feedback, and thermal conduction) and work done via gas compression, and use it to fit the observed temperature profiles by running Bayesian regressions. The results show that in all cases our model provides an acceptable fit at the 68% confidence level. To further validate this model we select nine clusters that have been observed with both Chandra (out to $\gtrsim 0.3r_{500}$) and Suzaku (out to $\gtrsim 1.5r_{500}$), fit their Chandra spectra with our model, and compare the extrapolation of the best-fits with the Suzaku measurements. We find that the model profiles agree with the Suzaku results very well in seven clusters. In the rest two clusters the difference between the model and observation is possibly caused by local thermal substructures. Our study also implies that for most of the clusters the assumption of hydrostatic equilibrium is safe out to at least $0.5r_{500}$, and the non-gravitational interactions between dark matter and its luminous counterpart is consistent with zero.