Search

In this work, we propose a modular approach for the Vision-Language Navigation (VLN) task by decomposing the problem into four sub-modules that use state-of-the-art Large Language Models (LLMs) and Vision-Language Models (VLMs) in a zero-shot setting. Given navigation instruction in natural language, we first prompt LLM to extract the landmarks and the order in which they are visited. Assuming the known model of the environment, we retrieve the top-k locations of the last landmark and generate $k$ path hypotheses from the starting location to the last landmark using the shortest path algorithm on the topological map of the environment. Each path hypothesis is represented by a sequence of panoramas. We then use dynamic programming to compute the alignment score between the sequence of panoramas and the sequence of landmark names, which match scores obtained from VLM. Finally, we compute the nDTW metric between the hypothesis that yields the highest alignment score to evaluate the path fidelity. We demonstrate superior performance compared to other approaches that use joint semantic maps like VLMaps \cite{vlmaps} on the complex R2R-Habitat \cite{r2r} instruction dataset and quantify in detail the effect of visual grounding on navigation performance.

Decoding visual images from brain activity has significant potential for advancing brain-computer interaction and enhancing the understanding of human perception. Recent approaches align the representation spaces of images and brain activity to enable visual decoding. In this paper, we introduce the use of human-aligned image encoders to map brain signals to images. We hypothesize that these models more effectively capture perceptual attributes associated with the rapid visual stimuli presentations commonly used in visual brain data recording experiments. Our empirical results support this hypothesis, demonstrating that this simple modification improves image retrieval accuracy by up to 21% compared to state-of-the-art methods. Comprehensive experiments confirm consistent performance improvements across diverse EEG architectures, image encoders, alignment methods, participants, and brain imaging modalities.

Training Large Language Models (LLMs) demand significant time and computational resources due to their large model sizes and optimizer states. To overcome these challenges, recent methods, such as BAdam, employ partial weight updates to enhance time and memory efficiency, though sometimes at the cost of performance. Others, like GaLore, focus on maintaining performance while optimizing memory usage through full parameter training, but may incur higher time complexity. By leveraging the low-rank structure of the gradient and the Grassmannian geometry, we propose SubTrack-Grad, a subspace tracking-based optimization method that efficiently tracks the evolving gradient subspace by incorporating estimation errors and previously identified subspaces. SubTrack-Grad delivers better or on-par results compared to GaLore, while significantly outperforming BAdam, which, despite being time-efficient, compromises performance. SubTrack-Grad reduces wall-time by up to 20.57% on GLUE tasks (15% average reduction) and up to 65% on SuperGLUE tasks (22% average reduction) compared to GaLore. Notably, for a 3B parameter model, GaLore incurred a substantial 157% increase in wall-time compared to full-rank training, whereas SubTrack-Grad exhibited a 31% increase, representing a 49% reduction in wall-time, while enjoying the same memory reductions as GaLore.

Enterprise ontology serves as a foundational framework for semantically comprehending the nature of organizations and the essential components that uphold their integrity. The systematic and conceptual understanding of organizations has garnered significant attention from researchers due to its pivotal role in various domains, including business modeling, enterprise architecture, business process management, context-aware systems, application development, interoperability across diverse systems and platforms, knowledge management, organizational learning and innovation, and conflict resolution within organizations. Achieving a consensus on the concepts related to the fundamental elements that constitute an organization is therefore critical. This study aims to conduct a comprehensive analysis and comparison of existing conceptual models of enterprises as documented in scholarly articles published over the past decade. We discuss the strengths and weaknesses of each model and introduce a robust framework for their evaluation. To facilitate this evaluation, we propose several pertinent criteria derived from established methodologies for assessing ontologies. Furthermore, we identify contemporary challenges and issues that have been overlooked in prior studies, offering insights and suggestions for future research directions in enterprise modeling. This article ultimately presents a roadmap for enhancing the systematic understanding of organizations through refined enterprise ontology frameworks.

Automatic Speech Recognition (ASR) systems have demonstrated remarkable performance across various applications. However, limited data and the unique language features of specific domains, such as low-resource languages, significantly degrade their performance and lead to higher Word Error Rates (WER). In this study, we propose Generative Error Correction via Retrieval-Augmented Generation (GEC-RAG), a novel approach designed to improve ASR accuracy for low-resource domains, like Persian. Our approach treats the ASR system as a black-box, a common practice in cloud-based services, and proposes a Retrieval-Augmented Generation (RAG) approach within the In-Context Learning (ICL) scheme to enhance the quality of ASR predictions. By constructing a knowledge base that pairs ASR predictions (1-best and 5-best hypotheses) with their corresponding ground truths, GEC-RAG retrieves lexically similar examples to the ASR transcription using the Term Frequency-Inverse Document Frequency (TF-IDF) measure. This process provides relevant error patterns of the system alongside the ASR transcription to the Generative Large Language Model (LLM), enabling targeted corrections. Our results demonstrate that this strategy significantly reduces WER in Persian and highlights a potential for domain adaptation and low-resource scenarios. This research underscores the effectiveness of using RAG in enhancing ASR systems without requiring direct model modification or fine-tuning, making it adaptable to any domain by simply updating the transcription knowledge base with domain-specific data., Comment: 6 pages

We investigate the role of quantum entanglement and coherence in suppressing radiation and explore its implications for dark matter. Using Dicke's framework, we demonstrate that entangled states in a gas at thermal equilibrium can lead to subradiance, trapping energy in dark quantum states and reducing radiation intensity. Applying this to the 21 cm line in dark matter halos, we find that quantum coherence renders the gas effectively dark. Moreover, entanglement results in a vanishing collision cross-section, consistent with the collisionless nature of dark matter observed in systems like the bullet cluster. We also show that absorption of incident radiation can exceed levels predicted by Beer's law, which holds only in the absence of coherence. These findings suggest that quantum entanglement and coherence may explain the non-luminous behavior of matter in dark matter halos, offering a novel perspective on dark matter and advancing the understanding of astrophysical radiative processes., Comment: Main text: 17 pages, Total: 33 pages, 7 figures

Fast radio bursts (FRBs), especially those from repeating sources, exhibit a rich variety of morphologies in their dynamic spectra (or waterfalls). Characterizing these morphologies and spectro-temporal properties is a key strategy in investigating the underlying unknown emission mechanism of FRBs. This type of analysis has been typically accomplished using two-dimensional Gaussian techniques and the autocorrelation function (ACF) of the waterfall. These techniques suffer from high uncertainties when applied to recently observed ultra-FRBs: FRBs that are only a few microseconds long. We present a technique that involves the tagging of per-channel arrival times of an FRB to perform sub-burst slope measurements. This technique leverages the number of frequency channels and can increase the precision of sub-burst slope measurements by several orders of magnitude, allowing it to be easily applied to ultra-FRBs and microshot forests. While scattering and dispersion remain important and often dominating sources of uncertainty in measurements, this technique provides an adaptable and firm foundation for obtaining spectro-temporal properties from all kinds of FRB morphologies. We present measurements using this technique of several hundred bursts across 12 repeating sources, including over 400 bursts from the repeating sources FRB 20121102A, FRB 20220912A, and FRB 20200120E, all of which exhibit microsecond-long FRBs, as well as 136 drift rates. In addition to retrieving the known relationship between sub-burst slope and duration, we explore other correlations between burst properties. We find that ultra-FRBs obey the sub-burst slope law along with longer duration bursts, and appear to form a distinct population in the duration-frequency relation., Comment: 17 pages, 9 figures. Submitted to MNRAS

Artificial intelligence (AI) has significantly advanced speech recognition applications. However, many existing neural network-based methods struggle with noise, reducing accuracy in real-world environments. This study addresses isolated spoken Persian digit recognition (zero to nine) under noisy conditions, particularly for phonetically similar numbers. A hybrid model combining residual convolutional neural networks and bidirectional gated recurrent units (BiGRU) is proposed, utilizing word units instead of phoneme units for speaker-independent recognition. The FARSDIGIT1 dataset, augmented with various approaches, is processed using Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction. Experimental results demonstrate the model's effectiveness, achieving 98.53%, 96.10%, and 95.92% accuracy on training, validation, and test sets, respectively. In noisy conditions, the proposed approach improves recognition by 26.88% over phoneme unit-based LSTM models and surpasses the Mel-scale Two Dimension Root Cepstrum Coefficients (MTDRCC) feature extraction technique along with MLP model (MTDRCC+MLP) by 7.61%., Comment: 6 pages, two columns, submitted to Pattern Recognition Letters

We present a mathematical analysis of the spectro-temporal properties of Fast Radio Bursts (FRBs), focusing on the distortions introduced by propagation effects such as scattering and inaccurate de-dispersion. By examining the impact of different scattering timescales and residual dispersion measures (DMs), both independently and in combination, we identify systematic trends in the sub-burst slope law as defined within the framework of the Triggered Relativistic Dynamical Model (TRDM). These effects primarily alter the measurements of the sub-burst slope and duration, thereby also modifying their correlations with other properties, such as central frequency and bandwidth. Our results show that scatter-induced temporal broadening affects duration more than slope, with weak to moderate scattering subtly modifying the sub-burst slope law and strong scattering causing significant deviations. Residual dispersion preferentially modifies the slope, further changing the trends predicted by the sub-burst slope law. Ultra-short bursts (or ultra-FRBs) emerge as particularly susceptible to these effects even at relatively high frequencies, underscoring the need for precise treatment of scattering and accurate dedispersion before performing analyses. Our findings emphasize the necessity for higher frequency observations (especially for ultra-FRBs) to improve the DM estimates as well as the measurements of spectro-temporal properties., Comment: 15 pages, 11 figures

Here, we present the outcomes from the second Large Language Model (LLM) Hackathon for Applications in Materials Science and Chemistry, which engaged participants across global hybrid locations, resulting in 34 team submissions. The submissions spanned seven key application areas and demonstrated the diverse utility of LLMs for applications in (1) molecular and material property prediction; (2) molecular and material design; (3) automation and novel interfaces; (4) scientific communication and education; (5) research data management and automation; (6) hypothesis generation and evaluation; and (7) knowledge extraction and reasoning from scientific literature. Each team submission is presented in a summary table with links to the code and as brief papers in the appendix. Beyond team results, we discuss the hackathon event and its hybrid format, which included physical hubs in Toronto, Montreal, San Francisco, Berlin, Lausanne, and Tokyo, alongside a global online hub to enable local and virtual collaboration. Overall, the event highlighted significant improvements in LLM capabilities since the previous year's hackathon, suggesting continued expansion of LLMs for applications in materials science and chemistry research. These outcomes demonstrate the dual utility of LLMs as both multipurpose models for diverse machine learning tasks and platforms for rapid prototyping custom applications in scientific research., Comment: Updating author information, the submission remains largely unchanged. 98 pages total

Spatially varying alignment of liquid crystals is essential for research and applications. One widely used method is based on the photopatterning of thin layers of azo-dye molecules, such as Brilliant Yellow (BY), that serve as an aligning substrate for a liquid crystal. In this study, we examine how photopatterning conditions, such as BY layer thickness (b), light intensity (I), irradiation dose, and age affect the alignment quality and the strength of the azimuthal surface anchoring. The azimuthal surface anchoring coefficient, W, is determined by analyzing the splitting of integer disclinations into half-integer disclinations at pre patterned substrates. The strongest anchoring is achieved for b in the range of 5-8 nm. W increases with the dose, and within the same dose, W increases with I. Aging of a non-irradiated BY coating above 15 days reduces W. Our study also demonstrates that sealed photopatterned cells filled with a conventional nematic preserve their alignment quality for up to four weeks, after which time W decreases. This work suggests the optimization pathways for photoalignment of nematic liquid crystals., Comment: 24 Pages, 7 Figures

DNA microarray gene-expression data has been widely used to identify cancerous gene signatures. Microarray can increase the accuracy of cancer diagnosis and prognosis. However, analyzing the large amount of gene expression data from microarray chips pose a challenge for current machine learning researches. One of the challenges lie within classification of healthy and cancerous tissues is high dimensionality of gene expressions. High dimensionality decreases the accuracy of the classification. This research aims to apply a hybrid model of Genetic Algorithm and Neural Network to overcome the problem during subset selection of informative genes. Whereby, a Genetic Algorithm (GA) reduced dimensionality during feature selection and then a Multi-Layer perceptron Neural Network (MLP) is applied to classify selected genes. The performance evaluated by considering to the accuracy and the number of selected genes. Experimental results show the proposed method suggested high accuracy and minimum number of selected genes in comparison with other machine learning algorithms.

This article addresses the gap in computational painting analysis focused on Persian miniature painting, a rich cultural and artistic heritage. It introduces a novel approach using Convolutional Neural Networks (CNN) to classify Persian miniatures from five schools: Herat, Tabriz-e Avval, Shiraz-e Avval, Tabriz-e Dovvom, and Qajar. The method achieves an average accuracy of over 91%. A meticulously curated dataset captures the distinct features of each school, with a patch-based CNN approach classifying image segments independently before merging results for enhanced accuracy. This research contributes significantly to digital art analysis, providing detailed insights into the dataset, CNN architecture, training, and validation processes. It highlights the potential for future advancements in automated art analysis, bridging machine learning, art history, and digital humanities, thereby aiding the preservation and understanding of Persian cultural heritage., Comment: 20 pages, submitted to journal

The freeze tag problem (FTP) aims to awaken a swarm of robots with one or more initial awake robots as soon as possible. Each awake robot must touch a sleeping robot to wake it up. Once a robot is awakened, it can assist in awakening other sleeping robots. We study this problem inside a polygonal domain and present approximation algorithms for it., Comment: 7 pages, 3 figures

The persistent challenge of bias in machine learning models necessitates robust solutions to ensure parity and equal treatment across diverse groups, particularly in classification tasks. Current methods for mitigating bias often result in information loss and an inadequate balance between accuracy and fairness. To address this, we propose a novel methodology grounded in bilevel optimization principles. Our deep learning-based approach concurrently optimizes for both accuracy and fairness objectives, and under certain assumptions, achieving proven Pareto optimal solutions while mitigating bias in the trained model. Theoretical analysis indicates that the upper bound on the loss incurred by this method is less than or equal to the loss of the Lagrangian approach, which involves adding a regularization term to the loss function. We demonstrate the efficacy of our model primarily on tabular datasets such as UCI Adult and Heritage Health. When benchmarked against state-of-the-art fairness methods, our model exhibits superior performance, advancing fairness-aware machine learning solutions and bridging the accuracy-fairness gap. The implementation of FairBiNN is available on https://github.com/yazdanimehdi/FairBiNN., Comment: Accepted to NeurIPS 2024

We investigate the weak semileptonic decay of the $ \Omega^{+}_{ccb} \rightarrow \Xi^{++}_{cc} ~{\ell}\bar\nu_{\ell}$, where a triply heavy baryon with spin 1/2 decays into the observed doubly heavy baryon with spin 1/2, using QCD sum rule method in all lepton channels. We compute the six relevant vector and axial vector form factors entering the low energy matrix elements in full theory. The invariant form factors are building blocks, using the fit faction of which in terms of $ q^2 $ in whole physical region, we calculate the exclusive widths in three lepton channels. Our predictions may help the present and future experiments in the course of their search for doubly and triply heavy baryons., Comment: 23 Pages, 8 Figures and 4 Tables

We present a comprehensive analysis of the periodic flares observed in the 6.7 GHz methanol transition in G22.356+0.066, utilizing the Maxwell-Bloch equations (MBEs) as a framework to model these phenomena. By solving the one-dimensional MBEs, we describe the behavior of both the quasi-steady-state maser and transient superradiance regimes. Our findings indicate that the observed periodic flares, with varying timescales across different velocities, are consistent with the characteristics of Dicke's superradiance, triggered by a common radiative pump in regions of varying inverted column densities. This work provides new insights into the physical processes governing variability in maser-hosting regions and underscores the significance of superradiance as a powerful radiation mechanism in astrophysical environments., Comment: 5 pages, 7 figures

The main objective of this research is to identify and classify the opinions of Persian-speaking Twitter users related to drought crises in Iran and subsequently develop a model for detecting these opinions on the platform. To achieve this, a model has been developed using machine learning and text mining methods to detect the opinions of Persian-speaking Twitter users regarding the drought issues in Iran. The statistical population for the research included 42,028 drought-related tweets posted over a one-year period. These tweets were extracted from Twitter using keywords related to the drought crises in Iran. Subsequently, a sample of 2,300 tweets was qualitatively analyzed, labeled, categorized, and examined. Next, a four-category classification of users` opinions regarding drought crises and Iranians' resilience to these crises was identified. Based on these four categories, a machine learning model based on logistic regression was trained to predict and detect various opinions in Twitter posts. The developed model exhibits an accuracy of 66.09% and an F-score of 60%, indicating that this model has good performance for detecting Iranian Twitter users' opinions regarding drought crises. The ability to detect opinions regarding drought crises on platforms like Twitter using machine learning methods can intelligently represent the resilience level of the Iranian society in the face of these crises, and inform policymakers in this area about changes in public opinion.

One fundamental task for NLP is to determine the similarity between two texts and evaluate the extent of their likeness. The previous methods for the Persian language have low accuracy and are unable to comprehend the structure and meaning of texts effectively. Additionally, these methods primarily focus on formal texts, but in real-world applications of text processing, there is a need for robust methods that can handle colloquial texts. This requires algorithms that consider the structure and significance of words based on context, rather than just the frequency of words. The lack of a proper dataset for this task in the Persian language makes it important to develop such algorithms and construct a dataset for Persian text. This paper introduces a new transformer-based model to measure semantic similarity between Persian informal short texts from social networks. In addition, a Persian dataset named FarSSiM has been constructed for this purpose, using real data from social networks and manually annotated and verified by a linguistic expert team. The proposed model involves training a large language model using the BERT architecture from scratch. This model, called FarSSiBERT, is pre-trained on approximately 104 million Persian informal short texts from social networks, making it one of a kind in the Persian language. Moreover, a novel specialized informal language tokenizer is provided that not only performs tokenization on formal texts well but also accurately identifies tokens that other Persian tokenizers are unable to recognize. It has been demonstrated that our proposed model outperforms ParsBERT, laBSE, and multilingual BERT in the Pearson and Spearman's coefficient criteria. Additionally, the pre-trained large language model has great potential for use in other NLP tasks on colloquial text and as a tokenizer for less-known informal words.

In this paper we construct a dissipative quintessential cosmic inflation. For this purpose, we add a multiplicative dissipative term in the standard quintessence field Lagrangian. We consider the specific form of dissipation as the time integral including the Hubble parameter and an arbitrary function that describes the dissipative properties of the quintessential scalar field. Inflation parameters and observables are calculated under slow-roll approximations and a detailed calculation of the cosmological perturbations is performed in this setup. We consider different forms of potentials and calculate the scalar spectral index and tensor-to-scalar ratio for a constant as well as variable dissipation function. To check the reliability of this model, a numerical analysis on the model parameters space is done in confrontation with recent observational data. By comparing the results with observational joint datasets at 68% and 95% confidence levels, we obtain some constraints on the model parameters space, specially the dissipation factor with e-folds numbers N = 55 and N = 60. As some specific results, we show that the power-law potential with a constant dissipation factor and N = 60 is mildly consistent with observational data in some restricted domains of the model parameter space with very small and negative dissipation factor and a negligible tensor-toscalar ratio. But this case with N = 55 is consistent with observation considerably. For power-law potential and variable dissipation factor as $Q = {\alpha}\phi^n$, the consistency with observation is also considerable with a reliable tensor-to-scalar ratio. The quadratic and quartic potentials with variable dissipation function as $Q = {\alpha}\phi^n$ are consistent with Planck2018 TT, TE, EE+lowE+lensing data at the 68% and 95% levels of confidence for some intervals of the parameter n., Comment: Accepted for publication in Physics of the Dark Universe, 37 pages, 14 figures, 8 tables

Self-supervised pretraining methods with masked prediction demonstrate remarkable within-dataset performance in skeleton-based action recognition. However, we show that, unlike contrastive learning approaches, they do not produce well-separated clusters. Additionally, these methods struggle with generalization in few-shot settings. To address these issues, we propose Self-supervised Tuning for 3D Action Recognition in Skeleton sequences (STARS). Specifically, STARS first uses a masked prediction stage using an encoder-decoder architecture. It then employs nearest-neighbor contrastive learning to partially tune the weights of the encoder, enhancing the formation of semantic clusters for different actions. By tuning the encoder for a few epochs, and without using hand-crafted data augmentations, STARS achieves state-of-the-art self-supervised results in various benchmarks, including NTU-60, NTU-120, and PKU-MMD. In addition, STARS exhibits significantly better results than masked prediction models in few-shot settings, where the model has not seen the actions throughout pretraining. Project page: https://soroushmehraban.github.io/stars/

Sign Language Recognition (SLR) is a fast-growing field that aims to fill the communication gaps between the hearing-impaired and people without hearing loss. Existing solutions for Persian Sign Language (PSL) are limited to word-level interpretations, underscoring the need for more advanced and comprehensive solutions. Moreover, previous work on other languages mainly focuses on manipulating the neural network architectures or hardware configurations instead of benefiting from the aggregated results of multiple models. In this paper, we introduce PenSLR, a glove-based sign language system consisting of an Inertial Measurement Unit (IMU) and five flexible sensors powered by a deep learning framework capable of predicting variable-length sequences. We achieve this in an end-to-end manner by leveraging the Connectionist Temporal Classification (CTC) loss function, eliminating the need for segmentation of input signals. To further enhance its capabilities, we propose a novel ensembling technique by leveraging a multiple sequence alignment algorithm known as Star Alignment. Furthermore, we introduce a new PSL dataset, including 16 PSL signs with more than 3000 time-series samples in total. We utilize this dataset to evaluate the performance of our system based on four word-level and sentence-level metrics. Our evaluations show that PenSLR achieves a remarkable word accuracy of 94.58% and 96.70% in subject-independent and subject-dependent setups, respectively. These achievements are attributable to our ensembling algorithm, which not only boosts the word-level performance by 0.51% and 1.32% in the respective scenarios but also yields significant enhancements of 1.46% and 4.00%, respectively, in sentence-level accuracy.

The ability to understand and reason about spatial relationships between objects in images is an important component of visual reasoning. This skill rests on the ability to recognize and localize objects of interest and determine their spatial relation. Early vision and language models (VLMs) have been shown to struggle to recognize spatial relations. We extend the previously released What'sUp dataset and propose a novel comprehensive evaluation for spatial relationship understanding that highlights the strengths and weaknesses of 27 different models. In addition to the VLMs evaluated in What'sUp, our extensive evaluation encompasses 3 classes of Multimodal LLMs (MLLMs) that vary in their parameter sizes (ranging from 7B to 110B), training/instruction-tuning methods, and visual resolution to benchmark their performances and scrutinize the scaling laws in this task., Comment: Accepted to NeurIPS 2024 Workshop on Compositional Learning

The recently discovered ferroelectric nematic (NF) liquid crystals (LC) have been reported to show an extraordinarily large value of the real part of the dielectric constant at low frequencies. However, it was argued by Clark et al in Physical Review Research 6, 013195 (2024) that what was measured was the capacitance of the insulating layer at LC/electrode surface and not that of the liquid crystal. Here we describe the results of dielectric spectroscopy measurements of an NF material in cells with variable thickness of the insulating layers. Our measurements quantitatively verify the model by Clark et al. Additionally, our measurements in cells with bare conducting indium tin oxide surface provide a crude estimate of epsilon perpendicular~100 in the NF phase., Comment: 17 pages, 6 figures, Presented in Workshop on Ferroelectric Nematic Liquid Crystals in Ljubljana