{"application": "software-mentions", "version": "0.9.0-SNAPSHOT", "revision": "54efbc6", "date": "2026-10-14T14:42+0000", "md5": "9A24EEF560D405B4A30B348AD567494A", "biblio": {"doi": "10.1109/csp66295.2025.00016", "title": "A Blockchain-Enhanced Reversible Watermarking Framework for End-to-End Data Traceability in Federated Learning Systems", "authors": "Reda Bellafqira, Chlo\u00e9 Berton, Gouenou Coatrieux"}, "mentions": [{"type": "software", "software-type": "software", "software-name": {"rawForm": "Alg", "normalizedForm": "Alg", "offsetStart": 0, "offsetEnd": 3}, "context": "Alg. 2 ensures perfect reconstruction of the original image through two key mechanisms: the bijective mapping in the histogram shifting operation, which guarantees reversibility of the embedding process, and the accurate handling of overflow cases.", "mentionContextAttributes": {"used": {"value": false, "score": 9.250640869140625e-05}, "created": {"value": false, "score": 5.2094459533691406e-05}, "shared": {"value": false, "score": 2.384185791015625e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9998483657836914}, "created": {"value": false, "score": 0.0013415813446044922}, "shared": {"value": false, "score": 3.5762786865234375e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG16 2", "normalizedForm": "VGG16 2", "offsetStart": 16, "offsetEnd": 23}, "context": "We utilized the VGG16 2 architecture as our base model for classification and performed comparative analysis between watermarked and non-watermarked datasets.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9999088048934937}, "created": {"value": false, "score": 0.3437265157699585}, "shared": {"value": false, "score": 1.1920928955078125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9999088048934937}, "created": {"value": false, "score": 0.3437265157699585}, "shared": {"value": false, "score": 1.1920928955078125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG16", "normalizedForm": "VGG16", "offsetStart": 26, "offsetEnd": 31}, "context": "We utilized a pre-trained VGG16 network fine-tuned for COVID-19 classification (Covid19, Normal and Viral Pneumonia), with training parameters set to 30 epochs, batch size of 32, Adam optimizer with learning rate 0.001, and the dataset split into 80% for training, 10% for testing, and 10% for validation.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 2.2530555725097656e-05}, "shared": {"value": false, "score": 4.76837158203125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 0.0021739602088928223}, "shared": {"value": false, "score": 4.76837158203125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG16", "normalizedForm": "VGG16", "offsetStart": 31, "offsetEnd": 36}, "context": "We trained two versions of the VGG16 model: i) Model trained on original (non-watermarked) dataset and ii) Model trained on watermarked dataset.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9998843669891357}, "created": {"value": false, "score": 0.0015419721603393555}, "shared": {"value": false, "score": 1.1920928955078125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 0.0021739602088928223}, "shared": {"value": false, "score": 4.76837158203125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG16", "normalizedForm": "VGG16", "offsetStart": 37, "offsetEnd": 42}, "context": "Testing on a medical dataset using a VGG16 model has demonstrated that our framework not only preserves model performance but shows slight improvement in accuracy.", "mentionContextAttributes": {"used": {"value": true, "score": 0.6715990900993347}, "created": {"value": false, "score": 0.0021739602088928223}, "shared": {"value": false, "score": 2.384185791015625e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 0.0021739602088928223}, "shared": {"value": false, "score": 4.76837158203125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "Alg", "normalizedForm": "Alg", "offsetStart": 50, "offsetEnd": 53}, "context": "The watermark embedding procedure is presented in Alg. 1.", "mentionContextAttributes": {"used": {"value": false, "score": 0.22761332988739014}, "created": {"value": false, "score": 8.404254913330078e-05}, "shared": {"value": false, "score": 3.5762786865234375e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9998483657836914}, "created": {"value": false, "score": 0.0013415813446044922}, "shared": {"value": false, "score": 3.5762786865234375e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "Alg", "normalizedForm": "Alg", "offsetStart": 77, "offsetEnd": 80}, "context": "Thus, to detect whether an image was logged in the blockchain, the following Alg. 3 is implemented.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9998483657836914}, "created": {"value": false, "score": 0.0013415813446044922}, "shared": {"value": false, "score": 1.1920928955078125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9998483657836914}, "created": {"value": false, "score": 0.0013415813446044922}, "shared": {"value": false, "score": 3.5762786865234375e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG-16", "normalizedForm": "VGG-16", "offsetStart": 94, "offsetEnd": 100}, "context": "We further show that watermarking does not interfere in the training and inference phase of a VGG-16 classification model for a Covid-19 medical database.", "mentionContextAttributes": {"used": {"value": false, "score": 0.0017316341400146484}, "created": {"value": false, "score": 0.0003292560577392578}, "shared": {"value": false, "score": 2.384185791015625e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.5945611000061035}, "created": {"value": false, "score": 0.0005785226821899414}, "shared": {"value": false, "score": 4.76837158203125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "Alg", "normalizedForm": "Alg", "offsetStart": 112, "offsetEnd": 115}, "context": "The predicted value P (y+ k h 2 , x+ kw 2 ) is computed by applying the kernel K to the neighborhood (step 6 in Alg.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9997754693031311}, "created": {"value": false, "score": 1.5616416931152344e-05}, "shared": {"value": false, "score": 2.384185791015625e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9998483657836914}, "created": {"value": false, "score": 0.0013415813446044922}, "shared": {"value": false, "score": 3.5762786865234375e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "VGG-16", "normalizedForm": "VGG-16", "offsetStart": 121, "offsetEnd": 127}, "context": "Furthermore, we show that the watermarking process does not adversely affect the training or inference capabilities of a VGG-16 classification model for COVID-19 diagnosis, with models trained on protected data maintaining their effectiveness on non-watermarked data.", "mentionContextAttributes": {"used": {"value": true, "score": 0.5945611000061035}, "created": {"value": false, "score": 0.0005785226821899414}, "shared": {"value": false, "score": 4.76837158203125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.5945611000061035}, "created": {"value": false, "score": 0.0005785226821899414}, "shared": {"value": false, "score": 4.76837158203125e-07}}}, {"type": "software", "software-type": "software", "software-name": {"rawForm": "Adam", "normalizedForm": "Adam", "offsetStart": 179, "offsetEnd": 183}, "context": "We utilized a pre-trained VGG16 network fine-tuned for COVID-19 classification (Covid19, Normal and Viral Pneumonia), with training parameters set to 30 epochs, batch size of 32, Adam optimizer with learning rate 0.001, and the dataset split into 80% for training, 10% for testing, and 10% for validation.", "mentionContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 2.2530555725097656e-05}, "shared": {"value": false, "score": 4.76837158203125e-07}}, "documentContextAttributes": {"used": {"value": true, "score": 0.9999370574951172}, "created": {"value": false, "score": 2.2530555725097656e-05}, "shared": {"value": false, "score": 4.76837158203125e-07}}}], "references": [], "runtime": 12160, "id": "9a6f42659f92bc42e55c72783fa433a1ce86e6b4", "metadata": {"id": "9a6f42659f92bc42e55c72783fa433a1ce86e6b4"}, "original_file_path": "../../projects/softcite-analysis-hal/data/grobid/xml/ABlockchainEnhancedReversibleWatermarkingFrameworkforEndtoEndDataTraceabilityinFederatedLearningSystems.grobid.tei.xml", "file_name": "ABlockchainEnhancedReversibleWatermarkingFrameworkforEndtoEndDataTraceabilityinFederatedLearningSystems.grobid.tei.xml"}