arXiv subjects

Amir Hajian

Publications and source records attributed to Amir Hajian.

At least 19 recordsLinked to original sources

arXiv2026

Real-Time Video Anomaly Detection Using YOLO Pose Estimation and CLIP-Based Semantic Scoring

We propose a lightweight two-stage framework for real-time video anomaly detection. The first stage employs YOLO v11n-pose to detect persons and extract seventeen skeletal keypoints in a single forward pass. The second stage encodes each cropped person region through CLIP ViT-B/32 and computes cosine similarity against predefined textual descriptions of anomalous behaviors. This architecture eliminates the need for optical flow, standalone pose estimators, and density-based scoring modules. Experiments on CUHK Avenue, ShanghaiTech Campus, and a custom indoor dataset collected at Chulalongkorn University demonstrate an end-to-end throughput of approximately 51 FPS on an NVIDIA Titan XP GPU, a 3.36x speedup over the multi-feature baseline, while maintaining frame-level AUROC values of 89.26%, 70.26%, and 84.13%, respectively.

cs.CV↗

arXiv2025

GraphiT: Efficient Node Classification on Text-Attributed Graphs with Prompt Optimized LLMs

The application of large language models (LLMs) to graph data has attracted a lot of attention recently. LLMs allow us to use deep contextual embeddings from pretrained models in text-attributed graphs, where shallow embeddings are often used for the text attributes of nodes. However, it is still challenging to efficiently encode the graph structure and features into a sequential form for use by LLMs. In addition, the performance of an LLM alone, is highly dependent on the structure of the input prompt, which limits their effectiveness as a reliable approach and often requires iterative manual adjustments that could be slow, tedious and difficult to replicate programmatically. In this paper, we propose GraphiT (Graphs in Text), a framework for encoding graphs into a textual format and optimizing LLM prompts for graph prediction tasks. Here we focus on node classification for text-attributed graphs. We encode the graph data for every node and its neighborhood into a concise text to enable LLMs to better utilize the information in the graph. We then further programmatically optimize the LLM prompts using the DSPy framework to automate this step and make it more efficient and reproducible. GraphiT outperforms our LLM-based baselines on three datasets and we show how the optimization step in GraphiT leads to measurably better results without manual prompt tweaking. We also demonstrated that our graph encoding approach is competitive to other graph encoding methods while being less expensive because it uses significantly less tokens for the same task.

cs.AI↗

arXiv2024

An Advanced Features Extraction Module for Remote Sensing Image Super-Resolution

In recent years, convolutional neural networks (CNNs) have achieved remarkable advancement in the field of remote sensing image super-resolution due to the complexity and variability of textures and structures in remote sensing images (RSIs), which often repeat in the same images but differ across others. Current deep learning-based super-resolution models focus less on high-frequency features, which leads to suboptimal performance in capturing contours, textures, and spatial information. State-of-the-art CNN-based methods now focus on the feature extraction of RSIs using attention mechanisms. However, these methods are still incapable of effectively identifying and utilizing key content attention signals in RSIs. To solve this problem, we proposed an advanced feature extraction module called Channel and Spatial Attention Feature Extraction (CSA-FE) for effectively extracting the features by using the channel and spatial attention incorporated with the standard vision transformer (ViT). The proposed method trained over the UCMerced dataset on scales 2, 3, and 4. The experimental results show that our proposed method helps the model focus on the specific channels and spatial locations containing high-frequency information so that the model can focus on relevant features and suppress irrelevant ones, which enhances the quality of super-resolved images. Our model achieved superior performance compared to various existing models.

eess.IV↗

arXiv2024

LongFin: A Multimodal Document Understanding Model for Long Financial Domain Documents

Document AI is a growing research field that focuses on the comprehension and extraction of information from scanned and digital documents to make everyday business operations more efficient. Numerous downstream tasks and datasets have been introduced to facilitate the training of AI models capable of parsing and extracting information from various document types such as receipts and scanned forms. Despite these advancements, both existing datasets and models fail to address critical challenges that arise in industrial contexts. Existing datasets primarily comprise short documents consisting of a single page, while existing models are constrained by a limited maximum length, often set at 512 tokens. Consequently, the practical application of these methods in financial services, where documents can span multiple pages, is severely impeded. To overcome these challenges, we introduce LongFin, a multimodal document AI model capable of encoding up to 4K tokens. We also propose the LongForms dataset, a comprehensive financial dataset that encapsulates several industrial challenges in financial documents. Through an extensive evaluation, we demonstrate the effectiveness of the LongFin model on the LongForms dataset, surpassing the performance of existing public models while maintaining comparable results on existing single-page benchmarks.

cs.CL↗

arXiv2016

A giant radio halo in a low-mass SZ-selected galaxy cluster: ACT-CL J0256.5+0006

We present the detection of a giant radio halo (GRH) in the Sunyaev-Zel'dovich (SZ)-selected merging galaxy cluster ACT-CL J0256.5+0006 ($z = 0.363$), observed with the Giant Metrewave Radio Telescope at 325 MHz and 610 MHz. We find this cluster to host a faint ($S_{610} = 5.6 \pm 1.4$ mJy) radio halo with an angular extent of 2.6 arcmin, corresponding to 0.8 Mpc at the cluster redshift, qualifying it as a GRH. J0256 is one of the lowest-mass systems, $M_{\rm 500,SZ} = (5.0 \pm 1.2) \times 10^{14} M_\odot$, found to host a GRH. We measure the GRH at lower significance at 325 MHz ($S_{325} = 10.3 \pm 5.3$ mJy), obtaining a spectral index measurement of $α^{610}_{325} = 1.0^{+0.7}_{-0.9}$. This result is consistent with the mean spectral index of the population of typical radio halos, $α= 1.2 \pm 0.2$. Adopting the latter value, we determine a 1.4 GHz radio power of $P_{1.4\text{GHz}} = (1.0 \pm 0.3) \times 10^{24}$ W Hz$^{-1}$, placing this cluster within the scatter of known scaling relations. Various lines of evidence, including the ICM morphology, suggest that ACT-CL J0256.5+0006 is composed of two subclusters. We determine a merger mass ratio of 7:4, and a line-of-sight velocity difference of $v_\perp = 1880 \pm 280$ km s$^{-1}$. We construct a simple merger model to infer relevant time-scales in the merger. From its location on the $P_{\rm 1.4GHz}{-}L_{\rm X}$ scaling relation, we infer that we observe ACT-CL J0256.5+0006 approximately 500 Myr before first core crossing.

astro-ph.GA↗

arXiv2015

Evidence of Lensing of the Cosmic Microwave Background by Dark Matter Halos

We present evidence of the gravitational lensing of the cosmic microwave background by $10^{13}$ solar mass dark matter halos. Lensing convergence maps from the Atacama Cosmology Telescope Polarimeter (ACTPol) are stacked at the positions of around 12,000 optically-selected CMASS galaxies from the SDSS-III/BOSS survey. The mean lensing signal is consistent with simulated dark matter halo profiles, and is favored over a null signal at 3.2 sigma significance. This result demonstrates the potential of microwave background lensing to probe the dark matter distribution in galaxy group and galaxy cluster halos.

astro-ph.CO↗

arXiv2015

SALT spectroscopic observations of galaxy clusters detected by ACT and a Type II quasar hosted by a brightest cluster galaxy

We present Southern African Large Telescope (SALT) follow-up observations of seven massive clusters detected by the Atacama Cosmology Telescope (ACT) on the celestial equator using the Sunyaev-Zel'dovich (SZ) effect. We conducted multi-object spectroscopic observations with the Robert Stobie Spectrograph in order to measure galaxy redshifts in each cluster field, determine the cluster line-of-sight velocity dispersions, and infer the cluster dynamical masses. We find that the clusters, which span the redshift range 0.3 < z < 0.55, range in mass from (5 -- 20) x 10$^{14}$ solar masses (M200c). Their masses, given their SZ signals, are similar to those of southern hemisphere ACT clusters previously observed using Gemini and the VLT. We note that the brightest cluster galaxy in one of the systems studied, ACT-CL J0320.4+0032 at z = 0.38, hosts a Type II quasar. Only a handful of such systems are currently known, and therefore ACT-CL J0320.4+0032 may be a rare example of a very massive halo in which quasar-mode feedback is actively taking place.

astro-ph.GA↗

arXiv2015

Machine Learning Etudes in Astrophysics: Selection Functions for Mock Cluster Catalogs

Making mock simulated catalogs is an important component of astrophysical data analysis. Selection criteria for observed astronomical objects are often too complicated to be derived from first principles. However the existence of an observed group of objects is a well-suited problem for machine learning classification. In this paper we use one-class classifiers to learn the properties of an observed catalog of clusters of galaxies from ROSAT and to pick clusters from mock simulations that resemble the observed ROSAT catalog. We show how this method can be used to study the cross-correlations of thermal Sunya'ev-Zeldovich signals with number density maps of X-ray selected cluster catalogs. The method reduces the bias due to hand-tuning the selection function and is readily scalable to large catalogs with a high-dimensional space of astrophysical features.

astro-ph.CO↗

arXiv2014

Testing Inflation with Large Scale Structure: Connecting Hopes with Reality

The statistics of primordial curvature fluctuations are our window into the period of inflation, where these fluctuations were generated. To date, the cosmic microwave background has been the dominant source of information about these perturbations. Large scale structure is however from where drastic improvements should originate. In this paper, we explain the theoretical motivations for pursuing such measurements and the challenges that lie ahead. In particular, we discuss and identify theoretical targets regarding the measurement of primordial non-Gaussianity. We argue that when quantified in terms of the local (equilateral) template amplitude $f_{\rm NL}^{\rm loc}$ ($f_{\rm NL}^{\rm eq}$), natural target levels of sensitivity are $Δf_{\rm NL}^{\rm loc, eq.} \simeq 1$. We highlight that such levels are within reach of future surveys by measuring 2-, 3- and 4-point statistics of the galaxy spatial distribution. This paper summarizes a workshop held at CITA (University of Toronto) on October 23-24, 2014.

astro-ph.CO↗

arXiv2014

First Measurement of the Cross-Correlation of CMB Lensing and Galaxy Lensing

We measure the cross-correlation of cosmic microwave background lensing convergence maps derived from Atacama Cosmology Telescope data with galaxy lensing convergence maps as measured by the Canada-France-Hawaii Telescope Stripe 82 Survey. The CMB-galaxy lensing cross power spectrum is measured for the first time with a significance of 4.2σ, which corresponds to a 12% constraint on the amplitude of density fluctuations at redshifts ~ 0.9. With upcoming improved lensing data, this novel type of measurement will become a powerful cosmological probe, providing a precise measurement of the mass distribution at intermediate redshifts and serving as a calibrator for systematic biases in weak lensing measurements.

astro-ph.CO↗

arXiv2014

The Atacama Cosmology Telescope: Lensing of CMB Temperature and Polarization Derived from Cosmic Infrared Background Cross-Correlation

We present a measurement of the gravitational lensing of the Cosmic Microwave Background (CMB) temperature and polarization fields obtained by cross-correlating the reconstructed convergence signal from the first season of ACTPol data at 146 GHz with Cosmic Infrared Background (CIB) fluctuations measured using the Planck satellite. Using an overlap area of 206 square degrees, we detect gravitational lensing of the CMB polarization by large-scale structure at a statistical significance of 4.5 sigma. Combining both CMB temperature and polarization data gives a lensing detection at 9.1 sigma significance. A B-mode polarization lensing signal is present with a significance of 3.2 sigma. We also present the first measurement of CMB lensing--CIB correlation at small scales corresponding to l > 2000. Null tests and systematic checks show that our results are not significantly biased by astrophysical or instrumental systematic effects, including Galactic dust. Fitting our measurements to the best-fit lensing-CIB cross power spectrum measured in Planck data, scaled by an amplitude A, gives A=1.02 +0.12/-0.18 (stat.) +/-0.06(syst.), consistent with the Planck results.

astro-ph.CO↗

arXiv2014

The Atacama Cosmology Telescope: A Measurement of the Thermal Sunyaev-Zel'dovich One-Point PDF

We present a measurement of the one-point probability distribution function (PDF) of the thermal Sunyaev-Zel'dovich (tSZ) decrement in the pixel temperature histogram of filtered 148 GHz sky maps from the Atacama Cosmology Telescope (ACT). The PDF includes the signal from all galaxy clusters in the map, including objects below the signal-to-noise threshold for individual detection, making it a particularly sensitive probe of the amplitude of matter density perturbations, $σ_8$. We use a combination of analytic halo model calculations and numerical simulations to compute the theoretical tSZ PDF and its covariance matrix, accounting for all noise sources and including relativistic corrections. From the measured ACT 148 GHz PDF alone, we find $σ_8 = 0.793 \pm 0.018$, with additional systematic errors of $\pm 0.017$ due to uncertainty in intracluster medium gas physics and $\pm 0.006$ due to uncertainty in infrared point source contamination. Using effectively the same data set, the statistical error here is a factor of two lower than that found in ACT's previous $σ_8$ determination based solely on the skewness of the tSZ signal. In future temperature maps with higher sensitivity, the tSZ PDF will break the degeneracy between intracluster medium gas physics and cosmological parameters.

astro-ph.CO↗

arXiv2014

A Measurement of the Millimeter Emission and the Sunyaev-Zel'dovich Effect Associated with Low-Frequency Radio Sources

We present a statistical analysis of the millimeter-wavelength properties of 1.4 GHz-selected sources and a detection of the Sunyaev-Zel'dovich (SZ) effect associated with the halos that host them. The Atacama Cosmology Telescope (ACT) has conducted a survey at 148 GHz, 218 GHz and 277 GHz along the celestial equator. Using samples of radio sources selected at 1.4 GHz from FIRST and NVSS, we measure the stacked 148, 218 and 277 GHz flux densities for sources with 1.4 GHz flux densities ranging from 5 to 200 mJy. At these flux densities, the radio source population is dominated by active galactic nuclei (AGN), with both steep and flat spectrum populations, which have combined radio-to-millimeter spectral indices ranging from 0.5 to 0.95, reflecting the prevalence of steep spectrum sources at high flux densities and the presence of flat spectrum sources at lower flux densities. The thermal SZ effect associated with the halos that host the AGN is detected at the 5$σ$ level through its spectral signature. When we compare the SZ effect with weak lensing measurements of radio galaxies, we find that the relation between the two is consistent with that measured by Planck for local bright galaxies. We present a detection of the SZ effect in some of the lowest mass halos (average $M_{200}\approx10^{13}$M$_{\odot}h_{70}^{-1}$) studied to date. This detection is particularly important in the context of galaxy evolution models, as it confirms that galaxies with radio AGN also typically support hot gaseous halos. With Herschel observations, we show that the SZ detection is not significantly contaminated by dust. We show that 5 mJy$<S_{1.4}<$200 mJy radio sources contribute $\ell(\ell+1)C_{\ell}/(2π)=0.37\pm0.03μ$K$^2$ to the angular power spectrum at $\ell=3000$ at 148 GHz, after accounting for the SZ effect associated with their host halos.

astro-ph.CO↗

arXiv2014

The Atacama Cosmology Telescope: CMB Polarization at $200<\ell<9000$

We report on measurements of the cosmic microwave background (CMB) and celestial polarization at 146 GHz made with the Atacama Cosmology Telescope Polarimeter (ACTPol) in its first three months of observing. Four regions of sky covering a total of 270 square degrees were mapped with an angular resolution of $1.3'$. The map noise levels in the four regions are between 11 and 17 $μ$K-arcmin. We present TT, TE, EE, TB, EB, and BB power spectra from three of these regions. The observed E-mode polarization power spectrum, displaying six acoustic peaks in the range $200<\ell<3000$, is an excellent fit to the prediction of the best-fit cosmological models from WMAP9+ACT and Planck data. The polarization power spectrum, which mainly reflects primordial plasma velocity perturbations, provides an independent determination of cosmological parameters consistent with those based on the temperature power spectrum, which results mostly from primordial density perturbations. We find that without masking any point sources in the EE data at $\ell<9000$, the Poisson tail of the EE power spectrum due to polarized point sources has an amplitude less than $2.4$ $μ$K$^2$ at $\ell = 3000$ at 95\% confidence. Finally, we report that the Crab Nebula, an important polarization calibration source at microwave frequencies, has 8.7\% polarization with an angle of $150.7^\circ \pm 0.6^\circ$ when smoothed with a $5'$ Gaussian beam.

astro-ph.CO↗

arXiv2014

Precision Epoch of Reionization studies with next-generation CMB experiments

Future arcminute resolution polarization data from ground-based Cosmic Microwave Background (CMB) observations can be used to estimate the contribution to the temperature power spectrum from the primary anisotropies and to uncover the signature of reionization near $\ell=1500$ in the small angular-scale temperature measurements. Our projections are based on combining expected small-scale E-mode polarization measurements from Advanced ACTPol in the range $300<\ell<3000$ with simulated temperature data from the full Planck mission in the low and intermediate $\ell$ region, $2<\ell<2000$. We show that the six basic cosmological parameters determined from this combination of data will predict the underlying primordial temperature spectrum at high multipoles to better than $1\%$ accuracy. Assuming an efficient cleaning from multi-frequency channels of most foregrounds in the temperature data, we investigate the sensitivity to the only residual secondary component, the kinematic Sunyaev-Zel'dovich (kSZ) term. The CMB polarization is used to break degeneracies between primordial and secondary terms present in temperature and, in effect, to remove from the temperature data all but the residual kSZ term. We estimate a $15 σ$ detection of the diffuse homogeneous kSZ signal from expected AdvACT temperature data at $\ell>1500$, leading to a measurement of the amplitude of matter density fluctuations, $σ_8$, at $1\%$ precision. Alternatively, by exploring the reionization signal encoded in the patchy kSZ measurements, we bound the time and duration of the reionization with $σ(z_{\rm re})=1.1$ and $σ(Δz_{\rm re})=0.2$. We find that these constraints degrade rapidly with large beam sizes, which highlights the importance of arcminute-scale resolution for future CMB surveys.

astro-ph.CO↗

arXiv2014

The co-evolution of black hole growth and star formation from a cross-correlation analysis between quasars and the cosmic infrared background

We present the first cross-correlation measurement between Sloan Digital Sky Survey (SDSS) Type 1 quasars and the cosmic infrared background (CIB) measured by Herschel. The distribution of the quasars at 0.15 =1.4) is $11.1^{+1.6}_{-1.4}$, $7.1^{+1.6}_{-1.3}$ and $3.6^{+1.4}_{-1.0}$ mJy at 250, 350 and 500 microns, respectively, while the mean sub-mm flux densities of the DR9 quasars ( =2.5) is $5.7^{+0.7}_{-0.6}$, $5.0^{+0.8}_{-0.7}$ and $1.8^{+0.5}_{-0.4}$ mJy. We find that the correlated sub-mm emission includes both the emission from satellite DSFGs in the same halo as the central quasar and the emission from DSFGs in separate halos (correlated with the quasar-hosting halo). The amplitude of the one-halo term is ~10 times smaller than the sub-mm emission of the quasars, implying the the satellites have a lower star-formation rate than the quasars. The satellite fraction for the DR7 quasars is $0.008^{+0.008}_{-0.005}$ and the host halo mass scale for the central and satellite quasars is $10^{12.36\pm0.87}$ M$_{\odot}$ and $10^{13.60\pm0.38}$ M$_{\odot}$, respectively. The satellite fraction of the DR9 quasars is $0.065^{+0.021}_{-0.031}$ and the host halo mass scale for the central and satellite quasars is $10^{12.29\pm0.62}$ M$_{\odot}$ and $10^{12.82\pm0.39}$ M$_{\odot}$, respectively. Thus, the typical halo environment of the SDSS Type 1 quasars is found to be similar to that of DSFGs, which supports the generally accepted view that dusty starburst and quasar activity are evolutionarily linked.

astro-ph.GA↗

arXiv2014

The Atacama Cosmology Telescope: Dusty Star-Forming Galaxies and Active Galactic Nuclei in the Southern Survey

We present a catalog of 191 extragalactic sources detected by the Atacama Cosmology Telescope (ACT) at 148 GHz and/or 218 GHz in the 2008 Southern survey. Flux densities span 14-1700 mJy, and we use source spectral indices derived using ACT-only data to divide our sources into two sub-populations: 167 radio galaxies powered by central active galactic nuclei (AGN), and 24 dusty star-forming galaxies (DSFGs). We cross-identify 97% of our sources (166 of the AGN and 19 of the DSFGs) with those in currently available catalogs. When combined with flux densities from the Australian Telescope 20 GHz survey and follow-up observations with the Australia Telescope Compact Array, the synchrotron-dominated population is seen to exhibit a steepening of the slope of the spectral energy distribution from 20 to 148 GHz, with the trend continuing to 218 GHz. The ACT dust-dominated source population has a median spectral index of 3.7+0.62-0.86, and includes both local galaxies and sources with redshifts as great as 5.6. Dusty sources with no counterpart in existing catalogs likely belong to a recently discovered subpopulation of DSFGs lensed by foreground galaxies or galaxy groups.

astro-ph.CO↗

arXiv2014

The Atacama Cosmology Telescope: Cross Correlation with Planck maps

We present the temperature power spectrum of the Cosmic Microwave Background obtained by cross-correlating maps from the Atacama Cosmology Telescope (ACT) at 148 and 218 GHz with maps from the Planck satellite at 143 and 217 GHz, in two overlapping regions covering 592 square degrees. We find excellent agreement between the two datasets at both frequencies, quantified using the variance of the residuals between the ACT power spectra and the ACTxPlanck cross-spectra. We use these cross-correlations to calibrate the ACT data at 148 and 218 GHz, to 0.7% and 2% precision respectively. We find no evidence for anisotropy in the calibration parameter. We compare the Planck 353 GHz power spectrum with the measured amplitudes of dust and cosmic infrared background (CIB) of ACT data at 148 and 218 GHz. We also compare planet and point source measurements from the two experiments.

astro-ph.CO↗