arXiv:2510.13137v2 Announce Type: replace 
Abstract: This study investigates the performance of 3D Convolutional Neural Networks (3D CNNs) and Long Short-Term Memory (LSTM) networks for real-time American Sign Language (ASL) recognition. Though 3D CNNs are good at spatiotemporal feature extraction from video sequences, LSTMs are optimized for modeling temporal dependencies in sequential data. We evaluate both architectures on a dataset containing 1,200 ASL signs across 50 classes, comparing their accuracy, computational efficiency, and latency under similar training conditions. Experimental results demonstrate that 3D CNNs achieve 92.4% recognition accuracy but require 3.2% more processing time per frame compared to LSTMs, which maintain 86.7% accuracy with significantly lower resource consumption. The hybrid 3D CNNLSTM model shows decent performance, which suggests that context-dependent architecture selection is crucial for practical implementation.This project provides professional benchmarks for developing assistive technologies, highlighting trade-offs between recognition precision and real-time operational requirements in edge computing environments.

تستكشف هذه الدراسة أداء الشبكات العصبية التلافيفية ثلاثية الأبعاد (3D CNN) والشبكات ذات الذاكرة طويلة المدى (LSTM) في التعرف على لغة الإشارة الأمريكية (ASL) في الوقت الحقيقي. تعتمد التقييمات على مجموعة بيانات تحتوي على 1200 إشارة ASL عبر 50 فئة، مع التركيز على الدقة والكفاءة الحاسوبية والزمن المستغرق. تظهر النتائج أن الشبكات ثلاثية الأبعاد تحقق دقة اعتراف بنسبة 92.4% ولكنها تتطلب وقت معالجة أكبر لكل إطار مقارنةً بالشبكات LSTM، التي تحافظ على دقة 86.7% مع استهلاك موارد أقل. يظهر النموذج الهجين أداءً جيدًا، مما يبرز أهمية اختيار الهيكل المناسب.

Este estudio investiga el rendimiento de las Redes Neuronales Convolucionales 3D (3D CNN) y las redes de Memoria a Largo Plazo (LSTM) para el reconocimiento en tiempo real de la Lengua de Señas Americana (ASL). La evaluación se basa en un conjunto de datos que contiene 1,200 signos ASL en 50 clases, centrándose en la precisión, la eficiencia computacional y la latencia. Los resultados muestran que las 3D CNN logran una precisión de reconocimiento del 92.4% pero requieren más tiempo de procesamiento por cuadro en comparación con las LSTM, que mantienen una precisión del 86.7% con un menor consu…

Cette étude examine la performance des réseaux de neurones convolutifs 3D (3D CNN) et des réseaux de mémoire à long terme (LSTM) pour la reconnaissance en temps réel de la langue des signes américaine (ASL). L'évaluation repose sur un ensemble de données de 1 200 signes ASL répartis sur 50 classes, en se concentrant sur la précision, l'efficacité computationnelle et la latence. Les résultats montrent que les 3D CNN atteignent une précision de reconnaissance de 92,4 % mais nécessitent plus de temps de traitement par image par rapport aux LSTM, qui maintiennent une précision de 86,7 % avec une c…

This study investigates the performance of 3D Convolutional Neural Networks (3D CNNs) and Long Short-Term Memory (LSTM) networks for real-time American Sign Language (ASL) recognition. The evaluation is based on a dataset of 1,200 ASL signs across 50 classes, focusing on accuracy, computational efficiency, and latency. Results show that 3D CNNs achieve 92.4% recognition accuracy but require more processing time per frame compared to LSTMs, which maintain 86.7% accuracy with lower resource consumption. A hybrid model demonstrates decent performance, highlighting the importance of architecture s…

Real-Time Sign Language to text Translation using Deep Learning: A Comparative study of LSTM and 3D CNN

arXiv:2601.07868v1 Announce Type: new 
Abstract: Dominant sequence models like the Transformer represent structure implicitly through dense attention weights, incurring quadratic complexity. We propose RewriteNets, a novel neural architecture built on an alternative paradigm: explicit, parallel string rewriting. Each layer in a RewriteNet contains a set of learnable rules. For each position in an input sequence, the layer performs four operations: (1) fuzzy matching of rule patterns, (2) conflict resolution via a differentiable assignment operator to select non-overlapping rewrites, (3) application of the chosen rules to replace input segments with output segments of potentially different lengths, and (4) propagation of untouched tokens. While the discrete assignment of rules is non-differentiable, we employ a straight-through Gumbel-Sinkhorn estimator, enabling stable end-to-end training. We evaluate RewriteNets on algorithmic, compositional, and string manipulation tasks, comparing them against strong LSTM and Transformer baselines. Results show that RewriteNets excel at tasks requiring systematic generalization (achieving 98.7% accuracy on the SCAN benchmark's length split) and are computationally more efficient than Transformers. We also provide an analysis of learned rules and an extensive ablation study, demonstrating that this architecture presents a promising direction for sequence modeling with explicit structural inductive biases.

يمثل تقديم RewriteNets تقدمًا كبيرًا في نمذجة التسلسل التوليدي، حيث يستخدم بنية جديدة تعتمد على إعادة كتابة السلاسل بشكل صريح ومتوازي بدلاً من الأوزان الكثيفة للاهتمام الموجودة في نماذج مثل Transformer. تتيح هذه الطريقة معالجة أكثر كفاءة من خلال إجراء مطابقة ضبابية، وحل النزاعات، ونشر الرموز بطريقة منظمة.

La introducción de RewriteNets marca un avance significativo en la modelización de secuencias generativas, utilizando una arquitectura novedosa que emplea la reescritura de cadenas explícita y paralela en lugar de los pesos de atención densos tradicionales que se encuentran en modelos como el Transformer. Este método permite un procesamiento más eficiente al realizar coincidencias difusas, resolución de conflictos y propagación de tokens de manera estructurada.

L'introduction des RewriteNets représente une avancée significative dans la modélisation des séquences génératives, utilisant une architecture novatrice qui emploie la réécriture de chaînes explicite et parallèle au lieu des poids d'attention denses traditionnels présents dans des modèles comme le Transformer. Cette méthode permet un traitement plus efficace en effectuant des correspondances floues, une résolution de conflits et une propagation de tokens de manière structurée.

The introduction of RewriteNets marks a significant advancement in generative sequence modeling, utilizing a novel architecture that employs explicit, parallel string rewriting instead of the traditional dense attention weights found in models like the Transformer. This method allows for more efficient processing by performing fuzzy matching, conflict resolution, and token propagation in a structured manner.

RewriteNets: End-to-End Trainable String-Rewriting for Generative Sequence Modeling

arXiv:2601.07951v1 Announce Type: new 
Abstract: Accurately forecasting long-term atmospheric variables remains a defining challenge in meteorological science due to the chaotic nature of atmospheric systems. Temperature data represents a complex superposition of deterministic cyclical climate forces and stochastic, short-term fluctuations. While planetary mechanics drive predictable seasonal periodicities, rapid meteorological changes such as thermal variations, pressure anomalies, and humidity shifts introduce nonlinear volatilities that defy simple extrapolation. Historically, the Seasonal Autoregressive Integrated Moving Average (SARIMA) model has been the standard for modeling historical weather data, prized for capturing linear seasonal trends. However, SARIMA operates under strict assumptions of stationarity, failing to capture abrupt, nonlinear transitions. This leads to systematic residual errors, manifesting as the under-prediction of sudden spikes or the over-smoothing of declines. Conversely, Deep Learning paradigms, specifically Long Short-Term Memory (LSTM) networks, demonstrate exceptional efficacy in handling intricate time-series data. By utilizing memory gates, LSTMs learn complex nonlinear dependencies. Yet, LSTMs face instability in open-loop forecasting; without ground truth feedback, minor deviations compound recursively, causing divergence. To resolve these limitations, we propose a Hybrid SARIMA-LSTM architecture. This framework employs a residual-learning strategy to decompose temperature into a predictable climate component and a nonlinear weather component. The SARIMA unit models the robust, long-term seasonal trend, while the LSTM is trained exclusively on the residuals the nonlinear errors SARIMA fails to capture. By fusing statistical stability with neural plasticity, this hybrid approach minimizes error propagation and enhances long-horizon accuracy.

تقدم دراسة جديدة نموذجًا هجينًا من SARIMA وLSTM يهدف إلى تحسين التنبؤ بالطقس المحلي من خلال نهج التعلم المتبقي، مما يعالج التحديات التي تطرحها الطبيعة الفوضوية للأنظمة الجوية. تكافح النماذج التقليدية مثل SARIMA مع الانتقالات المفاجئة وغير الخطية في بيانات درجات الحرارة، مما يؤدي إلى أخطاء منهجية في التنبؤات. يسعى النموذج الهجين إلى تعزيز الدقة من خلال دمج نقاط القوة في منهجيات SARIMA وLSTM.

Un nuevo estudio presenta un modelo híbrido SARIMA LSTM que busca mejorar la predicción del clima local mediante un enfoque de aprendizaje residual, abordando los desafíos que plantea la naturaleza caótica de los sistemas atmosféricos. Los modelos tradicionales como SARIMA tienen dificultades con las transiciones repentinas y no lineales en los datos de temperatura, lo que lleva a errores sistemáticos en las predicciones. El modelo híbrido busca aumentar la precisión al integrar las fortalezas de las metodologías SARIMA y LSTM.

Une nouvelle étude présente un modèle hybride SARIMA LSTM visant à améliorer les prévisions météorologiques locales grâce à une approche d'apprentissage résiduel, répondant aux défis posés par la nature chaotique des systèmes atmosphériques. Les modèles traditionnels comme SARIMA ont du mal avec les transitions soudaines et non linéaires des données de température, entraînant des erreurs systématiques dans les prévisions. Le modèle hybride cherche à améliorer la précision en intégrant les forces des méthodologies SARIMA et LSTM.

A new study presents a Hybrid SARIMA LSTM model aimed at improving local weather forecasting through a residual learning approach, addressing the challenges posed by the chaotic nature of atmospheric systems. Traditional models like SARIMA struggle with sudden, nonlinear transitions in temperature data, leading to systematic errors in predictions. The hybrid model seeks to enhance accuracy by integrating the strengths of both SARIMA and LSTM methodologies.

Real-Time Sign Language to text Translation using Deep Learning: A Comparative study of LSTM and 3D CNN

Was this article worth reading? Share it

LucidQuery AI

Humanize AI

Airparser

OpenL Translator

Synthesia

sync. labs

Ready to build your own newsroom?