Search

Current vision systems typically assign fixed-length representations to images, regardless of the information content. This contrasts with human intelligence - and even large language models - which allocate varying representational capacities based on entropy, context and familiarity. Inspired by this, we propose an approach to learn variable-length token representations for 2D images. Our encoder-decoder architecture recursively processes 2D image tokens, distilling them into 1D latent tokens over multiple iterations of recurrent rollouts. Each iteration refines the 2D tokens, updates the existing 1D latent tokens, and adaptively increases representational capacity by adding new tokens. This enables compression of images into a variable number of tokens, ranging from 32 to 256. We validate our tokenizer using reconstruction loss and FID metrics, demonstrating that token count aligns with image entropy, familiarity and downstream task requirements. Recurrent token processing with increasing representational capacity in each iteration shows signs of token specialization, revealing potential for object / part discovery., Comment: Code at: https://github.com/ShivamDuggal4/adaptive-length-tokenizer

The advent of 5G positioning techniques by 3GPP has unlocked possibilities for applications in public safety, vehicular systems, and location-based services. However, these applications demand accurate and reliable positioning performance, which has led to the proposal of newer positioning techniques. To further advance the research on these techniques, in this paper, we develop a 3GPP-compliant 5G positioning testbed, incorporating gNodeBs (gNBs) and User Equipment (UE). The testbed uses New Radio (NR) Positioning Reference Signals (PRS) transmitted by the gNB to generate Time of Arrival (TOA) estimates at the UE. We mathematically model the inter-gNB and UE-gNB time offsets affecting the TOA estimates and examine their impact on positioning performance. Additionally, we propose a calibration method for estimating these time offsets. Furthermore, we investigate the environmental impact on the TOA estimates. Our findings are based on our mathematical model and supported by experimental results., Comment: 8 pages, 9 figures, Accepted in ACM Wintech 2024

Mamba, a special case of the State Space Model, is gaining popularity as an alternative to template-based deep learning approaches in medical image analysis. While transformers are powerful architectures, they have drawbacks, including quadratic computational complexity and an inability to address long-range dependencies efficiently. This limitation affects the analysis of large and complex datasets in medical imaging, where there are many spatial and temporal relationships. In contrast, Mamba offers benefits that make it well-suited for medical image analysis. It has linear time complexity, which is a significant improvement over transformers. Mamba processes longer sequences without attention mechanisms, enabling faster inference and requiring less memory. Mamba also demonstrates strong performance in merging multimodal data, improving diagnosis accuracy and patient outcomes. The organization of this paper allows readers to appreciate the capabilities of Mamba in medical imaging step by step. We begin by defining core concepts of SSMs and models, including S4, S5, and S6, followed by an exploration of Mamba architectures such as pure Mamba, U-Net variants, and hybrid models with convolutional neural networks, transformers, and Graph Neural Networks. We also cover Mamba optimizations, techniques and adaptations, scanning, datasets, applications, experimental results, and conclude with its challenges and future directions in medical imaging. This review aims to demonstrate the transformative potential of Mamba in overcoming existing barriers within medical imaging while paving the way for innovative advancements in the field. A comprehensive list of Mamba architectures applied in the medical field, reviewed in this work, is available at Github.

The open world is inherently dynamic, characterized by ever-evolving concepts and distributions. Continual learning (CL) in this dynamic open-world environment presents a significant challenge in effectively generalizing to unseen test-time classes. To address this challenge, we introduce a new practical CL setting tailored for open-world visual representation learning. In this setting, subsequent data streams systematically introduce novel classes that are disjoint from those seen in previous training phases, while also remaining distinct from the unseen test classes. In response, we present Dynamic Prompt and Representation Learner (DPaRL), a simple yet effective Prompt-based CL (PCL) method. Our DPaRL learns to generate dynamic prompts for inference, as opposed to relying on a static prompt pool in previous PCL methods. In addition, DPaRL jointly learns dynamic prompt generation and discriminative representation at each training stage whereas prior PCL methods only refine the prompt learning throughout the process. Our experimental results demonstrate the superiority of our approach, surpassing state-of-the-art methods on well-established open-world image retrieval benchmarks by an average of 4.7% improvement in Recall@1 performance., Comment: ECCV 2024

Wireless positioning in Non-Line-of-Sight (NLOS) scenarios is highly challenging due to multipath, which leads to deterioration in the positioning estimate. This study reexamines electromagnetic field principles and applies them to wireless positioning, resulting in new techniques that enhance positioning accuracy in NLOS scenarios. Further, we use the proposed method to analyze a public safety scenario where it is essential to determine the position of at-risk individuals within buildings, emphasizing improving the Z-axis position estimate. Our analysis uses the Geometrical Theory of Diffraction (GTD) to provide important signal propagation insights and develop a new NLOS path model. Next, we use Fisher information to derive necessary and sufficient conditions for 3D positioning using our proposed positioning technique and finally to lower bound the possible 3D and z-axis positioning performance. On applying this positioning technique in a public safety scenario, we show that it is possible to greatly improve both 3D and Z-axis positioning performance by directly estimating NLOS path lengths.

Enhancing 3D and Z-axis positioning accuracy is crucial for effective rescue in indoor emergencies, ensuring safety for emergency responders and at-risk individuals. Additionally, reducing the dependence of a positioning system on fixed infrastructure is crucial, given its vulnerability to power failures and damage during emergencies. Further challenges from a signal propagation perspective include poor indoor signal coverage, multipath effects and the problem of Non-Line-OfSight (NLOS) measurement bias. In this study, we utilize the mobility provided by a rapidly deployable Uncrewed Aerial Vehicle (UAV) based wireless network to address these challenges. We recognize diffraction from window edges as a crucial signal propagation mechanism and employ the Geometrical Theory of Diffraction (GTD) to introduce a novel NLOS path length model. Using this path length model, we propose two different techniques to improve the indoor positioning performance for emergency scenarios., Comment: Accepted for publication in ICC 2024

Covariance representations are developed for the uniform distributions on the Euclidean spheres in terms of spherical gradients and Hessians. They are applied to derive a number of Sobolev type inequalities and to recover and refine the concentration of measure phenomenon, including second order concentration inequalities. A detail account is also given in the case of the circle, with a short overview of H\"offding's kernels and covariance identities in the class of periodic functions.

We start with a brief survey on H\"offding's kernels, its properties, related spectral decompositions, and discuss marginal distributions of H\"offding measures. In the second part of this note, one-dimensional covariance representations are considered over compactly supported probability distributions in the class of periodic smooth functions. H\"offding's kernels are used in the construction of mixing measures whose marginals are multiples of given probability distributions, leading to optimal kernels in periodic covariance representations., Comment: arXiv admin note: text overlap with arXiv:2403.19089

Inspired by the approach of Ivanisvili and Volberg towards functional inequalities for probability measures with strictly convex potentials, we investigate the relationship between curvature bounds in the sense of Bakry-Emery and local functional inequalities. We will show that not only is the earlier approach for strictly convex potentials extendable to Markov semigroups and simplified through use of the $\Gamma$-calculus, providing a consolidating machinery for obtaining functional inequalities new and old in this general setting, but that a converse also holds. Local inequalities obtained are proven equivalent to Bakry-Emery curvature. Moreover we will develop this technique for metric measure spaces satisfying the RCD condition, providing a unified approach to functional and isoperimetric inequalities in non-smooth spaces with a synthetic Ricci curvature bound. Finally, we are interested in commutation properties for semi-group operators on $\mathbb{R}^n$ in the absence of positive curvature, based on a local eigenvalue criteria.

The paper proves two results involving a pair (A,B) of P-biisometric or (m,P)-biisometric Hilbert-space operators for arbitrary positive integer m and positive operator P. It is shown that if A and B are power bounded and the pair (A,B) is (m,P)-biisometric for some m, then it is a P-biisometric pair. The important case when P is invertible is treated in detail. It is also shown that if (A,B) is P-biisometric, then there are biorthogonal sequences with respect to the inner product <.;.>_P= that have a shift-like behaviour with respect to this inner product., Comment: Linear and Multilinear Algebra, to appear (2024)

What does learning to model relationships between strings teach large language models (LLMs) about the visual world? We systematically evaluate LLMs' abilities to generate and recognize an assortment of visual concepts of increasing complexity and then demonstrate how a preliminary visual representation learning system can be trained using models of text. As language models lack the ability to consume or output visual information as pixels, we use code to represent images in our study. Although LLM-generated images do not look like natural images, results on image generation and the ability of models to correct these generated images indicate that precise modeling of strings can teach language models about numerous aspects of the visual world. Furthermore, experiments on self-supervised visual representation learning, utilizing images generated with text models, highlight the potential to train vision models capable of making semantic assessments of natural images using just LLMs.

The recent outbreak of the MPox virus has resulted in a tremendous increase in the usage of Twitter. Prior works in this area of research have primarily focused on the sentiment analysis and content analysis of these Tweets, and the few works that have focused on topic modeling have multiple limitations. This paper aims to address this research gap and makes two scientific contributions to this field. First, it presents the results of performing Topic Modeling on 601,432 Tweets about the 2022 Mpox outbreak that were posted on Twitter between 7 May 2022 and 3 March 2023. The results indicate that the conversations on Twitter related to Mpox during this time range may be broadly categorized into four distinct themes - Views and Perspectives about Mpox, Updates on Cases and Investigations about Mpox, Mpox and the LGBTQIA+ Community, and Mpox and COVID-19. Second, the paper presents the findings from the analysis of these Tweets. The results show that the theme that was most popular on Twitter (in terms of the number of Tweets posted) during this time range was Views and Perspectives about Mpox. This was followed by the theme of Mpox and the LGBTQIA+ Community, which was followed by the themes of Mpox and COVID-19 and Updates on Cases and Investigations about Mpox, respectively. Finally, a comparison with related studies in this area of research is also presented to highlight the novelty and significance of this research work.

The World Health Organization added Disease X to their shortlist of blueprint priority diseases to represent a hypothetical, unknown pathogen that could cause a future epidemic. During different virus outbreaks of the past, such as COVID-19, Influenza, Lyme Disease, and Zika virus, researchers from various disciplines utilized Google Trends to mine multimodal components of web behavior to study, investigate, and analyze the global awareness, preparedness, and response associated with these respective virus outbreaks. As the world prepares for Disease X, a dataset on web behavior related to Disease X would be crucial to contribute towards the timely advancement of research in this field. Furthermore, none of the prior works in this field have focused on the development of a dataset to compile relevant web behavior data, which would help to prepare for Disease X. To address these research challenges, this work presents a dataset of web behavior related to Disease X, which emerged from different geographic regions of the world, between February 2018 and August 2023. Specifically, this dataset presents the search interests related to Disease X from 94 geographic regions. The dataset was developed by collecting data using Google Trends. The relevant search interests for all these regions for each month in this time range are available in this dataset. This paper also discusses the compliance of this dataset with the FAIR principles of scientific data management. Finally, an analysis of this dataset is presented to uphold the applicability, relevance, and usefulness of this dataset for the investigation of different research questions in the interrelated fields of Big Data, Data Mining, Healthcare, Epidemiology, and Data Analysis with a specific focus on Disease X.

We present the first systematic search for UV signatures from radio source-driven AGN feedback in Compact Steep Spectrum (CSS) radio galaxies. Owing to their characteristic sub-galactic jets (1-20 kpc projected linear sizes), CSS hosts are excellent laboratories for probing galaxy scale feedback via jet-triggered star formation. The sample consists of 7 powerful CSS galaxies, and 2 galaxies host to radio sources >20 kpc as control, at low to intermediate redshifts (z<0.6). Our new HST images show extended UV continuum emission in 6/7 CSS galaxies; with 5 CSS hosts exhibiting UV knots co-spatial and aligned along the radio-jet axis. Young (<10 Myr), massive (>5 M$_\odot$) stellar populations are likely to be the dominant source of the blue excess emission in radio galaxies at these redshifts. Hence, the radio-aligned UV regions could be attributed to jet-induced starbursts. Lower near-UV SFRs compared to other indicators suggests low scattered AGN light contribution to the observed UV. Dust attenuation of UV emission appears unlikely from high internal extinction correction estimates in most sources. Comparison with evolutionary synthesis models shows that our observations are consistent with recent (~1-8 Myr old) star forming activity likely triggered by current or an earlier episode of radio emission, or by a confined radio source that has frustrated growth due to a dense environment. While follow-up spectroscopic and polarized light observations are needed to constrain the activity-related components in the observed UV, the detection of jet-induced star formation is a confirmation of an important prediction of the jet feedback paradigm., Comment: Submitted to ApJ

Our research aims to unify existing works' diverging opinions on how architectural components affect the adversarial robustness of CNNs. To accomplish our goal, we synthesize a suite of three generalizable robust architectural design principles: (a) optimal range for depth and width configurations, (b) preferring convolutional over patchify stem stage, and (c) robust residual block design through adopting squeeze and excitation blocks and non-parametric smooth activation functions. Through extensive experiments across a wide spectrum of dataset scales, adversarial training methods, model parameters, and network design spaces, our principles consistently and markedly improve AutoAttack accuracy: 1-3 percentage points (pp) on CIFAR-10 and CIFAR-100, and 4-9 pp on ImageNet. The code is publicly available at https://github.com/poloclub/robust-principles., Comment: Published at BMVC'23

It is known that supercyclicity implies strong stability. It is not known whether weak l-sequential supercyclicity implies weak stability. In this paper we prove that weak l-sequential supercyclicity implies weak quasistability. Corollaries concerning the characterisation of (i) weakly l-sequentially supercyclic vectors that are not (strongly) supercyclic, and (ii) weakly l-sequentially supercyclic isometries, are also proved.