Akka Zemmari
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Akka Zemmari has a total of 164 co-authors. Here are the top co-authors by number of shared publications:
- Jenny Benois-Pineau
University of Bordeaux
15 shared publications Mohamed Mosbah
Univ. Bordeaux, CNRS, Bordeaux INP, LaBRI, UMR 5800
13 shared publications- Vijay Laxmi
Malaviya National Institute of Technology
13 shared publications Jenny Benois-Pineau
10 shared publications- John Michael Robson
UMR CNRS 5800 351 cours de la Libération Université de Bordeaux -Bordeaux INP LaBRI 33405 Talence France
10 shared publications Manoj Singh Gaur
10 shared publications
Latest Publications
- 3 August 2026
- Published by Springer Nature
- p. 591-605
There Is More to Attention: Statistical Filtering Enhances Explanations in Vision Transformers
- 3 August 2026
- Published by Springer Nature
- p. 591-605
There Is More to Attention: Statistical Filtering Enhances Explanations in Vision Transformers
- Open Access
- 1 June 2026
- Published by Elsevier in SSRN Electronic Journal
Fast Online Learning for Non-Stationary Visual Streams via Moving-to-Data in Resource-Constrained Settings
Continual learning, and specifically Online Streaming Learning (OSL), has become critical for adapting machine learning models to continuous data streams. However, conventional approaches rely on backpropagation-based optimization. This incurs high computational costs, introduces latency, and may reduce model plasticity over prolonged updates, making these methods impractical for resource-constrained edge devices. To address this, we propose Move-to-Data (MTD), a lightweight hybrid learning framework designed for fast online adaptation. MTD replaces continuous gradient descent at the classifier level with rapid, geometry-driven analytical weight updates. To improve long-term learning and mitigate model drift, we introduce enhanced variants, including Entropy-based MTD (EMTD) and EMTD with Retargeting (EMTDR). These variants actively filter the data stream using entropy-based sample selection and apply sparse, conditional gradient corrections only when necessary. We evaluate MTD across continual learning benchmarks, domain adaptation tasks, and challenging real-world egocentric video datasets. Experiments on standard Vision Transformers demonstrate that our methods match or outperform state-of-the-art continual learning approaches while reducing training latency. Furthermore, evaluations on the edge-optimized MobileViT model confirm that our framework maintains comparable or superior accuracy. Compared to backpropagation-based baselines, our approach reduces total streaming training time by up to 800x, attains the lowest peak memory footprint, and updates only a small fraction of the model parameters when compared to feature space adapters. The code is available at https://gitub.u-bordeaux.fr/mayyar/family-of-mtd.
- 1 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFast Online Learning for Non-Stationary Visual Streams via Moving-to-Data in Resource-Constrained Settings
Continual learning, and specifically Online Streaming Learning (OSL), has become critical for adapting machine learning models to continuous data streams. However, conventional approaches rely on backpropagation-based optimization. This incurs high computational costs, introduces latency, and may reduce model plasticity over prolonged updates, making these methods impractical for resource-constrained edge devices. To address this, we propose Move-to-Data (MTD), a lightweight hybrid learning framework designed for fast online adaptation. MTD replaces continuous gradient descent at the classifier level with rapid, geometry-driven analytical weight updates. To improve long-term learning and mitigate model drift, we introduce enhanced variants, including Entropy-based MTD (EMTD) and EMTD with Retargeting (EMTDR). These variants actively filter the data stream using entropy-based sample selection and apply sparse, conditional gradient corrections only when necessary. We evaluate MTD across continual learning benchmarks, domain adaptation tasks, and challenging real-world egocentric video datasets. Experiments on standard Vision Transformers demonstrate that our methods match or outperform state-of-the-art continual learning approaches while reducing training latency. Furthermore, evaluations on the edge-optimized MobileViT model confirm that our framework maintains comparable or superior accuracy. Compared to backpropagation-based baselines, our approach reduces total streaming training time by up to 800x, attains the lowest peak memory footprint, and updates only a small fraction of the model parameters when compared to feature space adapters. The code is available at https://gitub.u-bordeaux.fr/mayyar/family-of-mtd.
- 1 May 2026
- Published by Springer Nature
- p. 246-262
Drug Repurposing for Rare Diseases: A Novel Knowledge Graph Explainable Approach to Identify Drug Candidates
- 1 May 2026
- Published by Springer Nature
- p. 246-262
Drug Repurposing for Rare Diseases: A Novel Knowledge Graph Explainable Approach to Identify Drug Candidates
- Open Access
FusWay: Multimodal hybrid fusion approach. Application to Railway Defect Detection
Multimodal fusion is a multimedia technique that has become popular in the wide range of tasks where image information is accompanied by a signal/audio. The latter may not convey highly semantic information, such as speech or music, but some measures such as audio signal recorded by mics in the goal to detect rail structure elements or defects. While classical detection approaches such as You Only Look Once (YOLO) family detectors can be efficiently deployed for defect detection on the image modality, the single modality approaches remain limited. They yield an overdetection in case of the appearance similar to normal structural elements. The paper proposes a new multimodal fusion architecture built on the basis of domain rules with YOLO and Vision transformer backbones. It integrates YOLOv8n for rapid object detection with a Vision Transformer (ViT) to combine feature maps extracted from multiple layers (7, 16, and 19) and synthesised audio representations for two defect classes: rail Rupture and Surface defect. Fusion is performed between audio and image. Experimental evaluation on a real-world railway dataset demonstrates that our multimodal fusion improves precision and overall accuracy by 0.2 points compared to the vision-only approach. Student's unpaired t-test also confirms statistical significance of differences in the mean accuracy.
Open AccessFusWay: Multimodal hybrid fusion approach. Application to Railway Defect Detection
Multimodal fusion is a multimedia technique that has become popular in the wide range of tasks where image information is accompanied by a signal/audio. The latter may not convey highly semantic information, such as speech or music, but some measures such as audio signal recorded by mics in the goal to detect rail structure elements or defects. While classical detection approaches such as You Only Look Once (YOLO) family detectors can be efficiently deployed for defect detection on the image modality, the single modality approaches remain limited. They yield an overdetection in case of the appearance similar to normal structural elements. The paper proposes a new multimodal fusion architecture built on the basis of domain rules with YOLO and Vision transformer backbones. It integrates YOLOv8n for rapid object detection with a Vision Transformer (ViT) to combine feature maps extracted from multiple layers (7, 16, and 19) and synthesised audio representations for two defect classes: rail Rupture and Surface defect. Fusion is performed between audio and image. Experimental evaluation on a real-world railway dataset demonstrates that our multimodal fusion improves precision and overall accuracy by 0.2 points compared to the vision-only approach. Student's unpaired t-test also confirms statistical significance of differences in the mean accuracy.
- 11 December 2024
- Published by Springer Nature
- p. 160-175
ESL: Explain to Improve Streaming Learning for Transformers
- 11 December 2024
- Published by Springer Nature
- p. 160-175
ESL: Explain to Improve Streaming Learning for Transformers
- 15 September 2024
- Published by Springer Nature in Network Modeling Analysis in Health Informatics and Bioinformatics
- Vol. 13 (1) , 1-25
Investigating the impact of body node coordinator position on communication reliability in wireless body area networks
- 15 September 2024
- Published by Springer Nature in Network Modeling Analysis in Health Informatics and Bioinformatics
- Vol. 13 (1) , 1-25
Investigating the impact of body node coordinator position on communication reliability in wireless body area networks
- 11 March 2024
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 83 (35) , 83339-83356
A hybrid transformer with domain adaptation using interpretability techniques for the application to the detection of risk situations
- 11 March 2024
- Published by Springer Nature in Multimedia Tools and Applications
- Vol. 83 (35) , 83339-83356
A hybrid transformer with domain adaptation using interpretability techniques for the application to the detection of risk situations
- Open Access
A Hybrid System for Defect Detection on Rail Lines through the Fusion of Object and Context Information
Defect detection on rail lines is essential for ensuring safe and efficient transportation. Current image analysis methods with deep neural networks (DNNs) for defect detection often focus on the defects themselves while ignoring the related context. In this work, we propose a fusion model that combines both a targeted defect search and a context analysis, which is seen as a multimodal fusion task. Our model performs rule-based decision-level fusion, merging the confidence scores of multiple individual models to classify rail-line defects. We call the model “hybrid” in the sense that it is composed of supervised learning components and rule-based fusion. We first propose an improvement to existing vision-based defect detection methods by incorporating a convolutional block attention module (CBAM) in the you only look once (YOLO) versions 5 (YOLOv5) and 8 (YOLOv8) architectures for the detection of defects and contextual image elements. This attention module is applied at different detection scales. The domain-knowledge rules are applied to fuse the detection results. Our method demonstrates improvements over baseline models in vision-based defect detection. The model is open for the integration of modalities other than an image, e.g., sound and accelerometer data.
Open AccessA Hybrid System for Defect Detection on Rail Lines through the Fusion of Object and Context Information
Defect detection on rail lines is essential for ensuring safe and efficient transportation. Current image analysis methods with deep neural networks (DNNs) for defect detection often focus on the defects themselves while ignoring the related context. In this work, we propose a fusion model that combines both a targeted defect search and a context analysis, which is seen as a multimodal fusion task. Our model performs rule-based decision-level fusion, merging the confidence scores of multiple individual models to classify rail-line defects. We call the model “hybrid” in the sense that it is composed of supervised learning components and rule-based fusion. We first propose an improvement to existing vision-based defect detection methods by incorporating a convolutional block attention module (CBAM) in the you only look once (YOLO) versions 5 (YOLOv5) and 8 (YOLOv8) architectures for the detection of defects and contextual image elements. This attention module is applied at different detection scales. The domain-knowledge rules are applied to fuse the detection results. Our method demonstrates improvements over baseline models in vision-based defect detection. The model is open for the integration of modalities other than an image, e.g., sound and accelerometer data.
- 29 January 2024
- Published by Springer Nature
- p. 117-131
IFI: Interpreting for Improving: A Multimodal Transformer with an Interpretability Technique for Recognition of Risk Events
- 29 January 2024
- Published by Springer Nature
- p. 117-131
IFI: Interpreting for Improving: A Multimodal Transformer with an Interpretability Technique for Recognition of Risk Events
- Open Access
- 20 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 21-27
Entropy-based Sampling for Streaming learning with Move-to-Data approach on Video
- 20 September 2023
- Published by Association for Computing Machinery (ACM)
- p. 21-27
Open AccessEntropy-based Sampling for Streaming learning with Move-to-Data approach on Video