arXiv subjects

Jiachen Jiang

Publications and source records attributed to Jiachen Jiang.

At least 37 records · Page 2Linked to original sources

arXiv2025

Modeling reflection spectra of super-Eddington X-ray sources

Relativistic reflection is a common feature in the X-ray observations of accreting compact objects. We present reflux, a new X-ray reflection model for spectral analysis of super-Eddington sources. We develop two relativistic reflection frameworks for super-Eddington accretion: a slim-disk model that self-consistently accounts for disk thickening and self-shadowing, and an optically thick wind model that treats reflection off a funnel-shaped outflow. The slim-disk model offers a geometry where the inner disk thickness is proportional to radius, becoming thicker as the mass accretion rate increases. The wind model measures the opening angle of the funnel, the wind speed, and wind acceleration radius. The slim-disk profile reduces the brightness of the blue horn in the Fe K emission line for a fixed emissivity and significantly changes the intensity profile for a lamppost geometry. The wind model shows a blue-shifted iron line due to high velocity outflows. Both models assume a spherically symmetric spacetime. We apply the wind model to the XMM-Newton spectrum of the tidal disruption event Swift J1644+57, where the Fe K profile is expected to be shaped by scattering in an outflowing funnel. We constrain the opening angle of the funnel and find a high velocity of the wind.

astro-ph.HE↗

arXiv2025

Improving Visual Discriminability of CLIP for Training-Free Open-Vocabulary Semantic Segmentation

Extending CLIP models to semantic segmentation remains challenging due to the misalignment between their image-level pre-training objectives and the pixel-level visual understanding required for dense prediction. While prior efforts have achieved encouraging results by reorganizing the final layer and features, they often inherit the global alignment bias of preceding layers, leading to suboptimal segmentation performance. In this work, we propose LHT-CLIP, a novel training-free framework that systematically exploits the visual discriminability of CLIP across layer, head, and token levels. Through comprehensive analysis, we reveal three key insights: (i) the final layers primarily strengthen image-text alignment with sacrifice of visual discriminability (e.g., last 3 layers in ViT-B/16 and 8 layers in ViT-L/14), partly due to the emergence of anomalous tokens; (ii) a subset of attention heads (e.g., 10 out of 144 in ViT-B/16) display consistently strong visual discriminability across datasets; (iii) abnormal tokens display sparse and consistent activation pattern compared to normal tokens. Based on these findings, we propose three complementary techniques: semantic-spatial reweighting, selective head enhancement, and abnormal token replacement to effectively restore visual discriminability and improve segmentation performance without any additional training, auxiliary pre-trained networks, or extensive hyperparameter tuning. Extensive experiments on 8 common semantic segmentation benchmarks demonstrate that LHT-CLIP achieves state-of-the-art performance across diverse scenarios, highlighting its effectiveness and practicality for real-world deployment.

cs.CV↗

arXiv2025

In-Context Learning for Non-Stationary MIMO Equalization

Channel equalization is fundamental for mitigating distortions such as frequency-selective fading and inter-symbol interference. Unlike standard supervised learning approaches that require costly retraining or fine-tuning for each new task, in-context learning (ICL) adapts to new channels at inference time with only a few examples. However, existing ICL-based equalizers are primarily developed for and evaluated on static channels within the context window. Indeed, to our knowledge, prior principled analyses and theoretical studies of ICL focus exclusively on the stationary setting, where the function remains fixed within the context. In this paper, we investigate the ability of ICL to address non-stationary problems through the lens of time-varying channel equalization. We employ a principled framework for designing efficient attention mechanisms with improved adaptivity in non-stationary tasks, leveraging algorithms from adaptive signal processing to guide better designs. For example, new attention variants can be derived from the Least Mean Square (LMS) adaptive algorithm, a Least Root Mean Square (LRMS) formulation for enhanced robustness, or multi-step gradient updates for improved long-term tracking. Experimental results demonstrate that ICL holds strong promise for non-stationary MIMO equalization, and that attention mechanisms inspired by classical adaptive algorithms can substantially enhance adaptability and performance in dynamic environments. Our findings may provide critical insights for developing next-generation wireless foundation models with stronger adaptability and robustness.

cs.LG↗

arXiv2025

From Compression to Expression: A Layerwise Analysis of In-Context Learning

In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks without weight updates by learning from demonstration sequences. While ICL shows strong empirical performance, its internal representational mechanisms are not yet well understood. In this work, we conduct a statistical geometric analysis of ICL representations to investigate how task-specific information is captured across layers. Our analysis reveals an intriguing phenomenon, which we term *Layerwise Compression-Expression*: early layers progressively produce compact and discriminative representations that encode task information from the input demonstrations, while later layers express these representations to incorporate the query and generate the prediction. This phenomenon is observed consistently across diverse tasks and a range of contemporary LLM architectures. We demonstrate that it has important implications for ICL performance -- improving with model size and the number of demonstrations -- and for robustness in the presence of noisy examples. To further understand the effect of the compact task representation, we propose a bias-variance decomposition and provide a theoretical analysis showing how attention mechanisms contribute to reducing both variance and bias, thereby enhancing performance as the number of demonstrations increases. Our findings reveal an intriguing layerwise dynamic in ICL, highlight how structured representations emerge within LLMs, and showcase that analyzing internal representations can facilitate a deeper understanding of model behavior.

cs.CL↗

arXiv2025

Probing the Strong Gravity Region of Black Holes with eXTP

We present the novel capabilities of the enhanced X-ray Timing and Polarimetry (eXTP) mission to study the strong gravity region around stellar-mass black holes in X-ray binary systems and supermassive black holes in active galactic nuclei. eXTP can combine X-ray spectral, timing, and polarimetric techniques to study the accretion process near black holes, measure black hole masses and spins, and test Einstein's theory of General Relativity in the strong field regime. We show how eXTP can improve the current measurements of black holes of existing X-ray missions and we discuss the scientific questions that can be addressed.

astro-ph.HE↗

arXiv2025

Testing and Combining Transient Spectral Classification Tools on 4MOST-like Blended Spectra

With the 4-meter Multi-Object Spectroscopic Telescope (4MOST) expected to provide an influx of transient spectra when it begins observations in early 2026 we consider the potential for real-time classification of these spectra. We investigate three extant spectroscopic transient classifiers: the Deep Automated Supernova and Host classifier (DASH), Next Generation SuperFit (NGSF) and SuperNova IDentification (SNID), with a focus on comparing the completeness and purity of the transient samples they produce. We manually simulate fibre losses critical for accurately determining host-contamination and use the 4MOST Exposure Time Calculator to produce realistic, 4MOST-like, host-galaxy contaminated spectra. We investigate the three classifiers individually and in all possible combinations. We find that a combination of DASH and NGSF can produce a SN Ia sample with a purity of 99.9% while successfully classifying 70% of SNe Ia. However, it struggles to classify non-SN Ia transients. We investigate photometric cuts to transient magnitude and the transient's fraction of total fibre flux, finding that both can be used to improve non-SN Ia transient classification completeness by 8--44% with SNe Ibc benefitting the most and superluminous (SL) SNe the least. Finally, we present an example classification plan for live classification and the predicted purities and completeness across five transient classes: Ia, Ibc, II, SL and non-SN transients. We find that it is possible to classify 75% of input spectra with >70% purity in all classes except non-SN transients. Precise values can be varied using different classifiers and photometric cuts to suit the needs of a given study.

astro-ph.SR↗

arXiv2025

X-ray reflection spectroscopy with improved calculations of the emission angle

The reflection spectrum produced by a cold medium illuminated by X-ray photons is not isotropic and its shape depends on the emission angle. In the reflection spectrum of an accretion disk of a black hole, the value of the emission angle changes over the disk and, in general, is different from the value of the inclination angle of the disk because of the light bending in the strong gravitational field of the black hole. Current reflection models make some approximations, as calculating a reflection spectrum taking the correct emission angle at every point of the disk into account would be too time-consuming and make the model too slow to analyze observations. In a recent paper, we showed that these approximations are unsuitable to fit high-quality black hole spectra expected from the next generation of X-ray missions. Here, we present a reflection model with improved calculations of the emission angle that solves this problem.

astro-ph.HE↗

arXiv2025

From Emergence to Control: Probing and Modulating Self-Reflection in Language Models

Self-reflection -- the ability of a large language model (LLM) to revisit, evaluate, and revise its own reasoning -- has recently emerged as a powerful behavior enabled by reinforcement learning with verifiable rewards (RLVR). While self-reflection correlates with improved reasoning accuracy, its origin and underlying mechanisms remain poorly understood. In this work, {\it we first show that self-reflection is not exclusive to RLVR fine-tuned models: it already emerges, albeit rarely, in pretrained models}. To probe this latent ability, we introduce Reflection-Inducing Probing, a method that injects reflection-triggering reasoning traces from fine-tuned models into pretrained models. This intervention raises self-reflection frequency of Qwen2.5 from 0.6\% to 18.6\%, revealing a hidden capacity for reflection. Moreover, our analysis of internal representations shows that both pretrained and fine-tuned models maintain hidden states that distinctly separate self-reflective from non-reflective contexts. Leveraging this observation, {\it we then construct a self-reflection vector, a direction in activation space associated with self-reflective reasoning}. By manipulating this vector, we enable bidirectional control over the self-reflective behavior for both pretrained and fine-tuned models. Experiments across multiple reasoning benchmarks show that enhancing these vectors improves reasoning performance by up to 12\%, while suppressing them reduces computational cost, providing a flexible mechanism to navigate the trade-off between reasoning quality and efficiency without requiring additional training. Our findings further our understanding of self-reflection and support a growing body of work showing that understanding model internals can enable precise behavioral control.

cs.LG↗

arXiv2025

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations

Task vectors offer a compelling mechanism for accelerating inference in in-context learning (ICL) by distilling task-specific information into a single, reusable representation. Despite their empirical success, the underlying principles governing their emergence and functionality remain unclear. This work proposes the Linear Combination Conjecture, positing that task vectors act as single in-context demonstrations formed through linear combinations of the original ones. We provide both theoretical and empirical support for this conjecture. First, we show that task vectors naturally emerge in linear transformers trained on triplet-formatted prompts through loss landscape analysis. Next, we predict the failure of task vectors on representing high-rank mappings and confirm this on practical LLMs. Our findings are further validated through saliency analyses and parameter visualization, suggesting an enhancement of task vectors by injecting multiple ones into few-shot prompts. Together, our results advance the understanding of task vectors and shed light on the mechanisms underlying ICL in transformer-based models.

cs.LG↗

arXiv2025

ProCrop: Learning Aesthetic Image Cropping from Professional Compositions

Image cropping is crucial for enhancing the visual appeal and narrative impact of photographs, yet existing rule-based and data-driven approaches often lack diversity or require annotated training data. We introduce ProCrop, a retrieval-based method that leverages professional photography to guide cropping decisions. By fusing features from professional photographs with those of the query image, ProCrop learns from professional compositions, significantly boosting performance. Additionally, we present a large-scale dataset of 242K weakly-annotated images, generated by out-painting professional images and iteratively refining diverse crop proposals. This composition-aware dataset generation offers diverse high-quality crop proposals guided by aesthetic principles and becomes the largest publicly available dataset for image cropping. Extensive experiments show that ProCrop significantly outperforms existing methods in both supervised and weakly-supervised settings. Notably, when trained on the new dataset, our ProCrop surpasses previous weakly-supervised methods and even matches fully supervised approaches. Both the code and dataset will be made publicly available to advance research in image aesthetics and composition analysis.

cs.CV↗

arXiv2025

Tracing Representation Progression: Analyzing and Enhancing Layer-Wise Similarity

Analyzing the similarity of internal representations has been an important technique for understanding the behavior of deep neural networks. Most existing methods for analyzing the similarity between representations of high dimensions, such as those based on Centered Kernel Alignment (CKA), rely on statistical properties of the representations for a set of data points. In this paper, we focus on transformer models and study the similarity of representations between the hidden layers of individual transformers. In this context, we show that a simple sample-wise cosine similarity metric is capable of capturing the similarity and aligns with the complicated CKA. Our experimental results on common transformers reveal that representations across layers are positively correlated, with similarity increasing when layers get closer. We provide a theoretical justification for this phenomenon under the geodesic curve assumption for the learned transformer. We then show that an increase in representation similarity implies an increase in predicted probability when directly applying the last-layer classifier to any hidden layer representation. We then propose an aligned training method to improve the effectiveness of shallow layer by enhancing the similarity between internal representations, with trained models that enjoy the following properties: (1) more early saturation events, (2) layer-wise accuracies monotonically increase and reveal the minimal depth needed for the given task, (3) when served as multi-exit models, they achieve on-par performance with standard multi-exit architectures which consist of additional classifiers designed for early exiting in shallow layers. To our knowledge, our work is the first to show that one common classifier is sufficient for multi-exit models. We conduct experiments on both vision and NLP tasks to demonstrate the performance of the proposed aligned training.

cs.AI↗

arXiv2025

Analyzing Fine-Grained Alignment and Enhancing Vision Understanding in Multimodal Language Models

Achieving better alignment between vision embeddings and Large Language Models (LLMs) is crucial for enhancing the abilities of Multimodal LLMs (MLLMs), particularly for recent models that rely on powerful pretrained vision encoders and LLMs. A common approach to connect the pretrained vision encoder and LLM is through a projector applied after the vision encoder. However, the projector is often trained to enable the LLM to generate captions, and hence the mechanism by which LLMs understand each vision token remains unclear. In this work, we first investigate the role of the projector in compressing vision embeddings and aligning them with word embeddings. We show that the projector significantly compresses visual information, removing redundant details while preserving essential elements necessary for the LLM to understand visual content. We then examine patch-level alignment -- the alignment between each vision patch and its corresponding semantic words -- and propose a *multi-semantic alignment hypothesis*. Our analysis indicates that the projector trained by caption loss improves patch-level alignment but only to a limited extent, resulting in weak and coarse alignment. To address this issue, we propose *patch-aligned training* to efficiently enhance patch-level alignment. Our experiments show that patch-aligned training (1) achieves stronger compression capability and improved patch-level alignment, enabling the MLLM to generate higher-quality captions, (2) improves the MLLM's performance by 16% on referring expression grounding tasks, 4% on question-answering tasks, and 3% on modern instruction-following benchmarks when using the same supervised fine-tuning (SFT) setting. The proposed method can be easily extended to other multimodal models.

cs.CV↗

arXiv2025

A rapid black hole spin or emission from the plunging region?

Emission from within the plunging region of black hole accretion flows has recently been detected in two X-ray binary systems. There is, furthermore, a possible discrepancy between the inferred spins of gravitational wave and electromagnetically detected black holes. Motivated by these two results we demonstrate, using theoretical calculations, numerical simulations and observational data, that the inclusion of emission from within the innermost stable circular orbit (ISCO) results in a black hole with a low spin producing a thermal continuum X-ray spectrum that mimics that produced by a much more rapidly rotating black hole surrounded by a disk with no emission from within the ISCO. We demonstrate this explicitly using the observed X-ray spectrum of a canonical soft-state high mass X-ray binary system M33 X-7. A vanishing ISCO temperature model requires a high spin $a_\bullet = 0.84\pm0.05$, as has been found previously in the literature. However, a disk around a Schwarzschild black hole can equally well (in fact slightly better) describe the data, provided that photons emitted from within the plunging region are included, and the ISCO stress is in line with that seen in numerical simulations of the accretion process. We then present an analysis of two further soft-state X-ray binaries (MAXI J1820+070 and MAXI J0637$-$430) which require the presence of intra-ISCO emission at high statistical significance. These two sources sit on the low-spin moderate-stress part of the degeneracy exhibited by M33 X-7, suggesting that when high quality data are available the high-spin low-stress region of parameter space is ruled out. We discuss how future advances in numerical simulations and data modelling will be essential to determining the spin of X-ray binary black holes which may well be systematically lower than current continuum fitting methods suggest.

astro-ph.HE↗

arXiv2025

Narrow-Line Seyfert 1 Galaxies Beyond the Local X-ray Universe: An X-ray spectral sample

Narrow-line Seyfert 1 AGNs (NLS1s) represent a unique stage in the black hole growth history, characterised by low black hole masses of approximately $10^{6}$-$10^{8}$ solar masses and around-Eddington accretion rates. X-ray studies of NLS1s have largely been confined to the local Universe ($z < 0.2$), while their broad-line counterparts and radio-loud quasars have been more extensively investigated at higher redshifts. In this work, we conducted an X-ray spectral analysis for 14 SDSS-observed NLS1s at $z\approx1$ in the eRASS1 catalogue. We found that all of their eROSITA observations agree with the expected rest-frame 2 keV monochromatic luminosity given their rest-frame 2500 angstrom monochromatic luminosity, further supporting evidence of AGN emission. Second, when fitted with a power-law model, most continuum spectra between 0.7-7 keV in their rest frames necessitate photon indices $Γ\gtrsim2.5$. Notably, the highest photon index of around 4.7 in one of our NLS1 AGNs hints at a significant contribution from soft excess emission. Finally, our analysis demonstrates that we can align the Eddington ratios with optical measurements by applying a correction factor between 10-120 to their X-ray luminosity. Although measurement uncertainty remains considerable, our findings suggest that assumptions for the standard geometrically thin accretion disc model made in previous estimations of this correction factor may not apply to near or super-Eddington NLS1 AGNs. Finally, we also compare this sample with extremely variable nearby NLS1s and other X-ray-weak AGNs, such as JWST-observed, broad-line AGNs at $z=5-6$, and underscores the importance of deeper X-ray surveys for more X-ray-weak NLS1s.

astro-ph.HE↗

arXiv2025

Extracting the X-ray reverberation response functions from the AGN light curves using an autoencoder

We study the X-ray reverberation in active galactic nuclei (AGN) using the variational autoencoder (VAE), which is a machine-learning algorithm widely used for signal processing and feature reconstruction. While the X-ray reverberation signatures that contain the information of the accretion disk and the X-ray emitting corona are commonly analyzed in the Fourier domain, this work aims to extract the reverberation response functions directly from the AGN light curves. The VAE is trained using the simulated light curves that contain the primary X-rays from the lamp-post corona varying its height and the corresponding reflection X-rays from the disk. We use progressively more realistic light-curve models, such as those that include the effects of disk-propagating fluctuations and random noises, to assess the ability of the VAE to reconstruct the response profiles. Interestingly, the VAE can recognize the reverberation patterns on the light curves, hence the coronal height can be predicted. We then deploy the VAE model to the XMM-Newton data of IRAS 13224-3809 and directly estimate, for the first time, the response functions of this source in various observations. The result reveals the corona changing its height between $3~r_{\rm g}$ and $20~r_{\rm g}$, which is correlated with the source luminosity and in line with previous literature. Finally, we discuss the advantages and limitations of this method.

astro-ph.HE↗

arXiv2024

X-ray spectra of black hole X-ray binaries with returning radiation

In the disk-corona model, the X-ray spectrum of a stellar-mass black hole in an X-ray binary is characterized by three components: a thermal component from a thin and cold accretion disk, a Comptonized component from a hot corona, and a reflection component produced by illumination of the cold disk by the hot corona. In this paper, we assume a lamppost corona and we improve previous calculations of the X-ray spectrum of black hole X-ray binaries. The reflection spectrum is produced by the direct radiation from the corona as well as by the returning radiation of the thermal and reflection components and is calculated considering the actual spectrum illuminating the disk. If we turn the corona off, the reflection spectrum is completely generated by the returning radiation of the thermal component, as it may happen for some sources in soft spectral states. After choosing the radial density profile of the accretion disk, the ionization parameter is calculated self-consistently at any radial coordinate of the disk from the illuminating X-ray flux and the local electron density. We show the predictions of our model in different regimes and we discuss its current limitations as well as the next steps to improve it.

astro-ph.HE↗

arXiv2024

Fifty Years After the Discovery of the First Stellar-Mass Black Hole: A Review of Cyg X-1

Around 50 years ago, the famous bet between Stephen Hawking and Kip Thorne on whether Cyg X-1 hosts a stellar-mass black hole became a well-known story in the history of black hole science. Today, Cyg X-1 is widely recognised as hosting a stellar-mass black hole with a mass of approximately 20 solar masses. With the advancement of X-ray telescopes, Cyg X-1 has become a prime laboratory for studies in stellar evolution, accretion physics, and high-energy plasma physics. In this review, we explore the latest results from X-ray observations of Cyg X-1, focusing on its implications for black hole spin, its role in stellar evolution, the geometry of the innermost accretion regions, and the plasma physics insights derived from its X-ray emissions. This review primarily focuses on Cyg X-1; however, the underlying physics applies to other black hole X-ray binaries and, to some extent, to AGNs.

astro-ph.HE↗

arXiv2024

Variable ionized disk wind in MAXI J1803-298 revealed by NICER

We present the results from the NICER observation data of MAXI J1803-298 across the entire 2021 outburst. In the intermediate and soft state, we detect significant absorption lines at $\sim 7.0$ keV and $\sim 6.7$ keV, arising from the X-ray disk wind outflowing with a velocity of hundreds of km per second along our line of sight. The fitting results from photoionized model suggest that the wind is driven by thermal pressure and the mass-loss rate is low. We find a clear transition for iron from predominantly H-like to predominantly He-like during the intermediate-to-soft state transition. Our results indicate this transition for iron is caused by the evolution of the illuminating spectrum and the slow change of the geometric properties of the disk wind together. The coexistence of disk wind and QPOs features in intermediate state is also reported. Our study makes MAXI J1803-298 the first source in which a transition from optical wind to X-ray wind is detected, offering new insights into the evolution of disk winds across an entire outburst and long-term coupling of accretion disks and mass outflows around accreting black holes.

astro-ph.HE↗