Search

Traditional machine learning approaches assume that data comes from a single generating mechanism, which may not hold for most real life data. In these cases, the single mechanism assumption can result in suboptimal performance. We introduce a clustering framework that eliminates this assumption by grouping the data according to the relations between the features and the target values and we obtain multiple separate models to learn different parts of the data. We further extend our framework to applications having streaming data where we produce outcomes using an ensemble of models. For this, the ensemble weights are updated based on the incoming data batches. We demonstrate the performance of our approach over the widely-studied real life datasets, showing significant improvements over the traditional single-model approaches.

Scaling the input image resolution is essential for enhancing the performance of Vision Language Models (VLMs), particularly in text-rich image understanding tasks. However, popular visual encoders such as ViTs become inefficient at high resolutions due to the large number of tokens and high encoding latency caused by stacked self-attention layers. At different operational resolutions, the vision encoder of a VLM can be optimized along two axes: reducing encoding latency and minimizing the number of visual tokens passed to the LLM, thereby lowering overall latency. Based on a comprehensive efficiency analysis of the interplay between image resolution, vision latency, token count, and LLM size, we introduce FastVLM, a model that achieves an optimized trade-off between latency, model size and accuracy. FastVLM incorporates FastViTHD, a novel hybrid vision encoder designed to output fewer tokens and significantly reduce encoding time for high-resolution images. Unlike previous methods, FastVLM achieves the optimal balance between visual token count and image resolution solely by scaling the input image, eliminating the need for additional token pruning and simplifying the model design. In the LLaVA-1.5 setup, FastVLM achieves 3.2$\times$ improvement in time-to-first-token (TTFT) while maintaining similar performance on VLM benchmarks compared to prior works. Compared to LLaVa-OneVision at the highest resolution (1152$\times$1152), FastVLM achieves comparable performance on key benchmarks like SeedBench and MMMU, using the same 0.5B LLM, but with 85$\times$ faster TTFT and a vision encoder that is 3.4$\times$ smaller.

Let $R\ $be an integral domain and $R^{\#}$ the set of all nonzero nonunits of $R.\ $For every elements $a,b\in R^{\#},$ we define $a\sim b$ if and only if $aR=bR,$ that is, $a$ and $b$ are associated elements. Suppose that $EC(R^{\#})$ is the set of all equivalence classes of $R^{\#}\ $according to $\sim$.$\ $Let $U_{a}=\{[b]\in EC(R^{\#}):b\ $divides $a\}$ for every $a\in R^{\#}.$ Then we prove that the family $\{U_{a}\}_{a\in R^{\#}}$ becomes a basis for a topology on $EC(R^{\#}).\ $This topology is called divisor topology of $R\ $and denoted by $D(R).\ $We investigate the connections between the algebraic properties of $R\ $and the topological properties of$\ D(R)$. In particular, we investigate the seperation axioms on $D(R)$, first and second countability axioms, connectivity and compactness on $D(R)$. We prove that for atomic domains $R,\ $the divisor topology $D(R)\ $is a Baire space. Also, we characterize valution domains $R$ in terms of nested property of $D(R).$ In the last section, we introduce a new topological proof of the infinitude of prime elements in a UFD and integers by using the topology $D(R)$., Comment: Comments welcome

In this paper, we associate a new topology to a nonzero unital module $M$ over a commutative $R$, which is called Golomb topology of the $R$-module $M$. Let $M\ $be an\ $R$-module and $B_{M}$ be the family of coprime cosets $\{m+N\}$ where $m\in M$ and $N\ $is a nonzero submodule of $M\ $such that $N+Rm=M$. We prove that if $M\ $is a meet irreducible multiplication module or $M\ $is a meet irreducible finitely generated module in which every maximal submodule is strongly irreducible, then $B_{M}\ $is the basis for a topology on $M\ $which is denoted by $\widetilde{G(M)}.$ In particular, the subspace topology on $M-\{0\}$ is called the Golomb topology of the $R$-module $M\ $and denoted by $G(M)$. We investigate the relations between topological properties of $G(M)\ $and algebraic properties of $M.\ $In particular, we characterize some important classes of modules such as simple modules, Jacobson semisimple modules in terms of Golomb topology., Comment: Comments welcome!

This study outlines a detailed investigation using CCD {\it UBV} and {\it Gaia} DR3 data sets of the two open clusters Ruprecht 1 (Rup-1) and Ruprecht 171 (Rup-171). Fundamental astrophysical parameters such as color excesses, photometric metallicities, ages, and isochrone distances were based on {\it UBV}-data analyses, whereas membership probability calculations, structural and astrophysical parameters, as well as the kinematic analyses were based on {\it Gaia} DR3-data. We identified 74 and 596 stars as the most probable cluster members with membership probabilities over 50\% for Rup-1 and Rup-171, respectively. The color excesses $E(B-V)$ were obtained as $0.166\pm0.022$ and $0.301\pm0.027$ mag for Rup-1 and Rup-171, respectively. Photometric metallicity analyses were performed by considering F-G type main-sequence member stars and found to be [Fe/H]=$-0.09\pm 0.16$ and [Fe/H]=$-0.20\pm 0.20$ dex for Rup-1 and Rup-171, respectively. Ages and distances were based on both {\it UBV} and {\it Gaia}-data analyses; according to isochrone-fitting these values were estimated to be $t=580\pm60$ Myr, $d=1469\pm57$ pc for Rup-1 and $t=2700\pm200$ Myr, $d=1509\pm69$ pc for Rup-171. The present-day mass function slope of Rup-1 was estimated as $1.26\pm0.32$ and Rup-171 as $1.53\pm1.49$. Galactic orbit integration analyses showed that both of the clusters might be formed outside the solar circle., Comment: 14 figures and 4 tables, article accepted for publication in Astronomische Nachrichten

This study explores the use of large language models (LLMs) to predict emotion intensity in Polish political texts, a resource-poor language context. The research compares the performance of several LLMs against a supervised model trained on an annotated corpus of 10,000 social media texts, evaluated for the intensity of emotions by expert judges. The findings indicate that while the supervised model generally outperforms LLMs, offering higher accuracy and lower variance, LLMs present a viable alternative, especially given the high costs associated with data annotation. The study highlights the potential of LLMs in low-resource language settings and underscores the need for further research on emotion intensity prediction and its application across different languages and continuous features. The implications suggest a nuanced decision-making process to choose the right approach to emotion prediction for researchers and practitioners based on resource availability and the specific requirements of their tasks., Comment: The Appendix is located at the very bottom of the manuscript

In this study, we attempt to model intuition and incorporate this formalism to improve the performance of the Convolutional Neural Networks. Despite decades of research, ambiguities persist on principles of intuition. Experimental psychology reveals many types of intuition, which depend on state of the human mind. We focus on visual intuition, useful for completing missing information during visual cognitive tasks. First, we set up a scenario to gradually decrease the amount of visual information in the images of a dataset to examine its impact on CNN accuracy. Then, we represent a model for visual intuition using Gestalt theory. The theory claims that humans derive a set of templates according to their subconscious experiences. When the brain decides that there is missing information in a scene, such as occlusion, it instantaneously completes the information by replacing the missing parts with the most similar ones. Based upon Gestalt theory, we model the visual intuition, in two layers. Details of these layers are provided throughout the paper. We use the MNIST data set to test the suggested intuition model for completing the missing information. Experiments show that the augmented CNN architecture provides higher performances compared to the classic models when using incomplete images., Comment: 7 pages, 5 figures, 1 table, submitted to 2024 IEEE International Conference on Image Processing (ICIP), did not get accepted

Generative-AI (GenAI), a novel technology capable of producing various types of outputs, including text, images, and videos, offers significant potential for wireless communications. This article introduces the concept of strategic demand-planning through demand-labeling, demand-shaping, and demand-rescheduling. Accordingly, GenAI is proposed as a powerful tool to facilitate demand-shaping in wireless networks. More specifically, GenAI is used to compress and convert the content of various types (e.g., from a higher bandwidth mode to a lower one, such as from a video to text), which subsequently enhances performance of wireless networks in various usage scenarios, such as cell-switching, user association and load balancing, interference management, as well as disasters and unusual gatherings. Therefore, GenAI can serve a function in saving energy and spectrum in wireless networks. With recent advancements in AI, including sophisticated algorithms like large language models and the development of more powerful hardware built exclusively for AI tasks, such as AI accelerators, the concept of demand-planning, particularly demand-shaping through GenAI, becomes increasingly relevant. Furthermore, recent efforts to make GenAI accessible on devices, such as user terminals, make the implementation of this concept even more straightforward and feasible.

Large language models (LLMs) are commonly trained on datasets consisting of fixed-length token sequences. These datasets are created by randomly concatenating documents of various lengths and then chunking them into sequences of a predetermined target length. However, this method of concatenation can lead to cross-document attention within a sequence, which is neither a desirable learning signal nor computationally efficient. Additionally, training on long sequences becomes computationally prohibitive due to the quadratic cost of attention. In this study, we introduce dataset decomposition, a novel variable sequence length training technique, to tackle these challenges. We decompose a dataset into a union of buckets, each containing sequences of the same size extracted from a unique document. During training, we use variable sequence length and batch size, sampling simultaneously from all buckets with a curriculum. In contrast to the concat-and-chunk baseline, which incurs a fixed attention cost at every step of training, our proposed method incurs a penalty proportional to the actual document lengths at each step, resulting in significant savings in training time. We train an 8k context-length 1B model at the same cost as a 2k context-length model trained with the baseline approach. Experiments on a web-scale corpus demonstrate that our approach significantly enhances performance on standard language evaluations and long-context benchmarks, reaching target accuracy 3x faster compared to the baseline. Our method not only enables efficient pretraining on long sequences but also scales effectively with dataset size. Lastly, we shed light on a critical yet less studied aspect of training large language models: the distribution and curriculum of sequence lengths, which results in a non-negligible difference in performance.

This study aims to introduce the cell load estimation problem of cell switching approaches in cellular networks specially-presented in a high-altitude platform station (HAPS)-assisted network. The problem arises from the fact that the traffic loads of sleeping base stations for the next time slot cannot be perfectly known, but they can rather be estimated, and any estimation error could result in divergence from the optimal decision, which subsequently affects the performance of energy efficiency. The traffic loads of the sleeping base stations for the next time slot are required because the switching decisions are made proactively in the current time slot. Two different Q-learning algorithms are developed; one is full-scale, focusing solely on the performance, while the other one is lightweight and addresses the computational cost. Results confirm that the estimation error is capable of changing cell switching decisions that yields performance divergence compared to no-error scenarios. Moreover, the developed Q-learning algorithms perform well since an insignificant difference (i.e., 0.3%) is observed between them and the optimum algorithm.

It is crucial to deploy temporary non-terrestrial networks (NTN) in disaster situations where terrestrial networks are no longer operable. Deploying uncrewed aerial vehicle base stations (UAV-BSs) can provide a radio access network (RAN); however, the backhaul link may also be damaged and unserviceable in such disaster conditions. In this regard, high-altitude platform stations (HAPS) spark attention as they can be deployed as super macro base stations (SMBS) and data centers. Therefore, in this study, we investigate a three-layer heterogeneous network with different topologies to prolong the lifespan of the temporary network by using UAV-BSs for RAN services and HAPS-SMBS as a backhaul. Furthermore, a two-layer clustering algorithm is proposed to handle the UAV-BS ad-hoc networking effectively.

In this study, we investigate the fundamental astrophysical parameters of the old open cluster NGC 188 through two complementary methods: isochron-fitting and spectral energy distribution (SED) analysis. Using photometric, astrometric, and spectroscopic data from the Gaia Data Release 3, we identify 868 most likely member stars with membership probabilities $P \geq 0.5$. The mean proper-motion components and trigonometric parallaxes of the cluster are derived as ($\mu_{\alpha}\cos \delta$, $\mu_{\delta}$) = (-$2.314 \pm 0.002$, -$1.022 \pm 0.002$) mas yr$^{-1}$ and $\varpi = 0.550 \pm 0.023$, respectively. From this initial selection of high probable member stars, we proceed with the determination of astrophysical parameters using the isochron-fitting method. Simultaneously estimating the colour excess, distance, and age of the cluster, we employee PARSEC isochrones to observational data on Gaia based colour-magnitude diagrams. These findings were obtained as $E(G_{BP}-G_{RP})=0.066\pm 0.012$ mag, $d=1806 \pm21$ pc, and $t=7.65 \pm 1.00$ Gyr, respectively. Additionally, we identify and detected 19 previously confirmed blue straggler stars within NGC 188. Subsequently, we performed SED analyses for 412 out of the 868 cluster members. We obtained colour excess, distance and age of the cluster as $E(B-V)=0.034\pm 0.030$ mag, $d=1854\pm 148$ pc, and $t=7.78\pm 0.23$ Gyr, respectively. The analysis of member stars was revealed patterns of extinction in the $V$-band, with higher values of A(V) observed in the lower right quadrant of the cluster. By comparing our results of SED analysis with models of stellar evolution, particularly in terms of temperature and surface gravity, we confirm agreement with theoretical predictions. This comprehensive investigation sheds light on the astrophysical properties of NGC 188, contributing to our understanding of stellar evolution within open clusters., Comment: 25 pages, 14 figures and 5 tables, accepted for publication in Physics and Astronomy Reports

In this article, we introduce the notion of regular fusible modules. Let $R$ be a ring with an identity and $M$ an $R$-module. An element $0\neq m\in M$ is said to be regular fusible if there exists $r\in R$, a non zero-divisor of $M$, such that $mr$ can be written as the sum of a torsion element and a torsion free element in $M$. $M$ is called regular fusible if every nonzero element of $M$ is regular fusible. We characterize regular fusible modules in terms of fusible modules. In addition, we show that a regular fusible module over a right duo ring is reduced and nonsingular. Moreover, we study the regular fusible property under Cartesian product, trivial extension ring, and module of a fractions. Also, we characterize division rings in terms of fusible modules.

This study investigates the integration of a high altitude platform station (HAPS), a non-terrestrial network (NTN) node, into the cell-switching paradigm for energy saving. By doing so, the sustainability and ubiquitous connectivity targets can be achieved. Besides, a delay-aware approach is also adopted, where the delay profiles of users are respected in such a way that we attempt to meet the latency requirements of users with a best-effort strategy. To this end, a novel, simple, and lightweight Q-learning algorithm is designed to address the cell-switching optimization problem. During the simulation campaigns, different interference scenarios and delay situations between base stations are examined in terms of energy consumption and quality-of-service (QoS), and the results confirm the efficacy of the proposed Q-learning algorithm.