Periodic Reporting for period 2 - GRACE (Global Response Against Child Exploitation)
Período documentado: 2021-12-01 hasta 2023-11-30
The growth in online child sexual exploitation and abuse (CSEAM) material is a significant challenge for European law enforcement agencies(LEA). Referrals of CSEAM exceed the capacity of LEAs to respond in a practical and timely manner. To safeguard victims, prosecute offenders and limit the spread of CSEAM, LEAs need a next-generation AI-powered investigative platform.
GRACE aims to equip European law enforcement agencies with advanced analytical and investigative capabilities to respond to the spread of online child sexual exploitation material. At the heart of the project, GRACE has three core concepts.
1- Address the volume and analyse the content of online CSEAM through technological innovations;
2.- Provide genuine operational value to LEAs in their investigation of online CSEAM; and
3.- Impact at the strategic and policy level in the harmonisation of EU-wide responses to CSEA.
GRACE will apply proven techniques in Machine Learning to the referral and analysis process while embracing the technical, ethical and legal challenges unique to fighting CSEA. GRACE will leverage resources already in place at Europol and the nine European Union Member State LEAs within the consortium. The goal for GRACE is to attempt to provide results early, frequently and flexibly, prioritising easy wins in the research plan (e.g. deduplication).
Unique to GRACE is the development and application of a Federated Learning approach to the challenge of optimising analysis and information flow in a privacy-aware and security-sensitive manner. GRACE will enable cooperation between LEAs in improving their capabilities while harnessing their experiential knowledge.
By the end of the Action, the GRACE project delivered a unique Collaborative Platform for the investigation and management of CSEM cases. It consists of a robust data ingestion pipeline, that follows the NCMEC referral composition but also provides a general API to integrate local data. It supports the analysis of multimedia files by 30 low level enrichment AI module (in the audio, video, image and text modalities), 11 analysis modules related to actionable intelligence (i.e. cross-matching, prioritisation and geo-localisation), and 1 targeted online search module. The GRACE Collaborative Platform provides a unique space for CSEM case management and investigation, with possibility of exchanging the analytical results among distinct entities at MS level.
Besides the Collaborative Platform, the GRACE project developed and featured three enabling capabilities, i.e. Federated Learning, Chain of Custody and Watermarking. These tackle important aspects in the practical life of law enforcement investigations, such as data segregation across the community, auditing of user activity on the platform and guaranteeing authenticity of information to investigative and judicial authorities. That is, aspects enforcing an auditable and trustworthy platform that can operate within the LE environment.
GRACE aims to equip European law enforcement agencies with advanced analytical and investigative capabilities to respond to the spread of online child sexual exploitation material. At the heart of the project, GRACE has three core concepts.
1- Address the volume and analyse the content of online CSEAM through technological innovations;
2.- Provide genuine operational value to LEAs in their investigation of online CSEAM; and
3.- Impact at the strategic and policy level in the harmonisation of EU-wide responses to CSEA.
GRACE will apply proven techniques in Machine Learning to the referral and analysis process while embracing the technical, ethical and legal challenges unique to fighting CSEA. GRACE will leverage resources already in place at Europol and the nine European Union Member State LEAs within the consortium. The goal for GRACE is to attempt to provide results early, frequently and flexibly, prioritising easy wins in the research plan (e.g. deduplication).
Unique to GRACE is the development and application of a Federated Learning approach to the challenge of optimising analysis and information flow in a privacy-aware and security-sensitive manner. GRACE will enable cooperation between LEAs in improving their capabilities while harnessing their experiential knowledge.
By the end of the Action, the GRACE project delivered a unique Collaborative Platform for the investigation and management of CSEM cases. It consists of a robust data ingestion pipeline, that follows the NCMEC referral composition but also provides a general API to integrate local data. It supports the analysis of multimedia files by 30 low level enrichment AI module (in the audio, video, image and text modalities), 11 analysis modules related to actionable intelligence (i.e. cross-matching, prioritisation and geo-localisation), and 1 targeted online search module. The GRACE Collaborative Platform provides a unique space for CSEM case management and investigation, with possibility of exchanging the analytical results among distinct entities at MS level.
Besides the Collaborative Platform, the GRACE project developed and featured three enabling capabilities, i.e. Federated Learning, Chain of Custody and Watermarking. These tackle important aspects in the practical life of law enforcement investigations, such as data segregation across the community, auditing of user activity on the platform and guaranteeing authenticity of information to investigative and judicial authorities. That is, aspects enforcing an auditable and trustworthy platform that can operate within the LE environment.
Throughout the duration of the action, the consortium focused on:
• Organise co-creation and cross-collaboration activities to advance on technological understanding among practitioners
• Collect end user requirements and elaborate systems architecture
• Develop unified categorisation and classification of the investigative information, i.e. a common GRACE ontology
• Support referrals from relevant NGOs and OSP/s through NCMEC reports
• Develop a variety of audio, video and text analytics modules to extract investigative relevant leads from multimedia sources;
• Develop a higher knowledge actionable intelligence tools to infer investigative value from the available report data, such as cross-matching, relevance classification, geo-localisation and prioritisation;
• Set up the essential prerequisites for the development of a Federated Learning platform, and its final testing covering image/video, audio and text analysis modalities.
• Develop the Collaboration platform front-ends and functionalities;
• Multiple rounds of operational testing through pilot activities, following the evaluation of the GRACE modules at different stages
• Large size annotation campaign on multi-valued image classification task; Preparation of text annotation campaign.
• Extended analysis of the available legislation and national frameworks to drive the GRACE developments and draft ways for cross-border collaboration;
• Elaborate training material and serious games.
• Participate at international fora dedicated to discussions relevant to the use of AI and fighting CSA
• Organise of a GRACE early Demonstration session (June 2023) and Final Conference (November 2023) in order to disseminate the project KERs towards an international CSEA community
GRACE has accomplished a tremendous amount across technology development, standardisation, policy, guidelines and training. This resulted in a set of 55 Key Exploitable results, spread among 8 different areas:
• Scientific Publications
• Research Datasets
• CSEM Taxonomy and Ontology
• GRACE system
• GRACE Technological components - divided into over 50 individual modules
• Ethical and Legal Framework and Guidelines
• Policy Recommendations, and
• Training Materials.
A detailed booklet on the GRACE KERs can be downloaded from the following link
https://grace-fct.eu/sites/default/files/2023-12/GRACE%20Key%20Exploitable%20Results%20%288mb%29.pdf(se abrirá en una nueva ventana)
Following the GA, all generated results are available to EU institutions, bodies, offices and agencies as well as EU Member States’ national authorities, on a royalty-free basis for non-commercial and non-competitive use, with the purpose of tackling the fight against CSEM.
• Organise co-creation and cross-collaboration activities to advance on technological understanding among practitioners
• Collect end user requirements and elaborate systems architecture
• Develop unified categorisation and classification of the investigative information, i.e. a common GRACE ontology
• Support referrals from relevant NGOs and OSP/s through NCMEC reports
• Develop a variety of audio, video and text analytics modules to extract investigative relevant leads from multimedia sources;
• Develop a higher knowledge actionable intelligence tools to infer investigative value from the available report data, such as cross-matching, relevance classification, geo-localisation and prioritisation;
• Set up the essential prerequisites for the development of a Federated Learning platform, and its final testing covering image/video, audio and text analysis modalities.
• Develop the Collaboration platform front-ends and functionalities;
• Multiple rounds of operational testing through pilot activities, following the evaluation of the GRACE modules at different stages
• Large size annotation campaign on multi-valued image classification task; Preparation of text annotation campaign.
• Extended analysis of the available legislation and national frameworks to drive the GRACE developments and draft ways for cross-border collaboration;
• Elaborate training material and serious games.
• Participate at international fora dedicated to discussions relevant to the use of AI and fighting CSA
• Organise of a GRACE early Demonstration session (June 2023) and Final Conference (November 2023) in order to disseminate the project KERs towards an international CSEA community
GRACE has accomplished a tremendous amount across technology development, standardisation, policy, guidelines and training. This resulted in a set of 55 Key Exploitable results, spread among 8 different areas:
• Scientific Publications
• Research Datasets
• CSEM Taxonomy and Ontology
• GRACE system
• GRACE Technological components - divided into over 50 individual modules
• Ethical and Legal Framework and Guidelines
• Policy Recommendations, and
• Training Materials.
A detailed booklet on the GRACE KERs can be downloaded from the following link
https://grace-fct.eu/sites/default/files/2023-12/GRACE%20Key%20Exploitable%20Results%20%288mb%29.pdf(se abrirá en una nueva ventana)
Following the GA, all generated results are available to EU institutions, bodies, offices and agencies as well as EU Member States’ national authorities, on a royalty-free basis for non-commercial and non-competitive use, with the purpose of tackling the fight against CSEM.
GRACE developed an innovative platform that paves the road to the novel interaction between EUROPOL and MS LEAs, that is aligned with the new Europol's mandate on fighting cross-border crime and strengthening the cooperation between EU MS. On a long term, the GRACE project will provide a reference framework for the future policies and the mandate of the new establishing EU Centre.
GRACE will drive the adoption of cutting edge technologies to the peculiarities of the CSEM fight domain. It developed a wide register of novel forensic analysis tools based ML technologies for rapid inspection of audio-visual and textual material, and contributing to the research on CSEAM-specific content analysis, classification, geo-localisation, evidence-graph creation and knowledge mining, case prioritisation and predictive trend analysis.
Central to the GRACE project is the adoption of a Federated Learning (FL) platform that is being designed and tested to train or tune the corresponding ML techniques on real data dispersed across MS LEAs. This novel approach will increase the capability of adapting the analytics to CSEM domain, while inherently implementing high standard principles of data privacy and security.
The principal innovation is in supporting the analysis of new CSEM material, that is supported through a novel Collaborative platform and more than 40 analytics tools. The innovation and progress beyond state of the art has been demonstrated for several of the tools through scientific publications.
Along with the technical work, GRACE also elaborated practical guidelines for First Responders and feedback to policies in the field of CSEM fight that shall provide insights for possible debate at the policy level related to the EU strategy for an effective fight against CSA.
GRACE will drive the adoption of cutting edge technologies to the peculiarities of the CSEM fight domain. It developed a wide register of novel forensic analysis tools based ML technologies for rapid inspection of audio-visual and textual material, and contributing to the research on CSEAM-specific content analysis, classification, geo-localisation, evidence-graph creation and knowledge mining, case prioritisation and predictive trend analysis.
Central to the GRACE project is the adoption of a Federated Learning (FL) platform that is being designed and tested to train or tune the corresponding ML techniques on real data dispersed across MS LEAs. This novel approach will increase the capability of adapting the analytics to CSEM domain, while inherently implementing high standard principles of data privacy and security.
The principal innovation is in supporting the analysis of new CSEM material, that is supported through a novel Collaborative platform and more than 40 analytics tools. The innovation and progress beyond state of the art has been demonstrated for several of the tools through scientific publications.
Along with the technical work, GRACE also elaborated practical guidelines for First Responders and feedback to policies in the field of CSEM fight that shall provide insights for possible debate at the policy level related to the EU strategy for an effective fight against CSA.