Ant Digital Technologies

🇭🇰 We’re proud to join the Hong Kong Monetary Authority (HKMA) and Cyberport Hong Kong as a technology partner in the second cohort of the GenAI Sandbox, supporting the industry’s move from exploring AI’s potential to implementing responsible and secure AI applications. 🧑💻 Through the Sandbox, Ant Digital Technologies will work with financial institutions — including Fubon Bank (Hong Kong) — leveraging our Agentar AI Agent Platform and mPaaS LUI module to enhance personalized and trusted mobile banking experiences. We will also apply our ZOLOZ AI solutions to strengthen identity fraud detection and defense. 🤝 We share HKMA’s commitment to governed innovation and will continue collaborating with partners across banking, technology, and academia to build a resilient AI ecosystem for the financial industry. #AntDigitalTechnologies #ResponsibleAI #FinTech https://lnkd.in/gec427QC

Thrilled to join the 10th anniversary edition of Hong Kong FinTech Week x StartmeupHK Festival 2025. Excited to connect, exchange, and co-create the next wave of responsible AI and digital transformation with global leaders in Hong Kong!

🚀 Congratulations to Hong Kong FinTech Week on its remarkable 10th anniversary! We look forward to joining global innovators in shaping the future of #FinTech, #AI, and #Sustainability.

🎉 This has been an important week for Ant Digital Technologies' technical progress！ 🏆 Our data analytics agent Agentar-SQL achieved top scores on the globally recognized BIRD-Bench, an evaluation for natural language to SQL conversion on real-world, production-scale databases. 🧠 Agentar-SQL is built on our proprietary large SQL model and trained with GSPO (Group Sequence Policy Optimization) reinforcement learning. 🧑💻 Our team continues to focus on practical, domain-driven AI—making analytics accessible through natural language while preserving rigor and reliability. It’s one step forward in applying large models to real enterprise data challenges. #AntDigitalTechnologies #BetterAIBetterWorld

🎙️We are proud to share that two privacy-preserving ML breakthroughs from Ant Digital Technologies have been accepted by ACM CCS 2025 and IEEE TDSC, two of the most prestigious venues in cybersecurity and trustworthy computing. 🐒 Gibbon, a secure GBDT training framework, integrating a series of innovative co-designs between training algorithms and advanced cryptography: Optimizes training algorithm to eliminate expensive cryptographic divisions, introduces a novel “linear + sine” approximation for the sigmoid, and employs a highly efficient matrix-vector multiplication protocol—delivering 2~4× faster secure training with less bandwidth, while keeping model accuracy. 📂 Privacy-preserving decision tree/graph inference from homomorphic lookup table: Instead of doing thousands of costly encrypted bit-by-bit comparisons, we pre-build homomorphically encrypted lookup tables and just “pull the answers.” That slashes communication by up to three orders of magnitude and is the first to privately support all test types (threshold / equality / set membership) plus scorecards. ⚙️ These techniques are already integrated into Ant’s privacy-preserving computing product matrix (FAIR for trusted data circulation platform, Morse for digital marketing platform, encrypted middleware, and LLM privacy suite), enabling secure and high-performance credit scoring, fraud detection, and precesion marketing across institutions. No more choosing between speed and confidentiality, institutions can have both now. 📚We appreciate the recognition from these two top-tier venues. Going forward, Ant Digital Technologies will continue to deepen core privacy-preserving computing technologies, promote the secure and efficient circulation of data elements, and contribute to the high-quality development of the digital economy. #AntDigitalTechnology #Research #AIForGood

🎙️Ant Digital Technologies, in collaboration with Nanyang Technological University, is proud to introduce LaV-CoT (Language-aware Visual Chain-of-Thought), a groundbreaking framework designed to tackle the challenges of multilingual visual question answering (mVQA) by unifying structured reasoning with language awareness, delivering state-of-the-art performance and elevates how AI handles complex multimodal tasks across diverse document and image contexts. 🧠 LaV-CoT stands out with its innovative four-stage flow—text extraction with spatial summaries, language identification, object and layout grounding, and concise logical reasoning - —supported by an automated multilingual data pipeline plus a two-stage training regimen (supervised fine-tuning and reinforcement optimization). This combination drives a 9.5% accuracy uplift over comparable open-source baselines and outperforms larger proprietary models on challenging non-English tasks. 🔐 LaV-CoT’s seamless integration into Ant Digital Technologies’s intelligent document processing product RealDoc strengthens multilingual document understanding for high-value scenarios such as visa processing, insurance claims, financial compliance, and other cross-border workflows. 🔎 Language identification and consistency control, the framework accurately detects document languages and delivers reliable cross-lingual comprehension. visual-text alignment via bounding-box summaries and spatial object-level image descriptions, plus interpretable step-by-step logical reasoning improve trust, auditability and deployment readiness, accelerating next-gen AI solutions across our global ecosystem. 📖 Full paper in the comment. #AntDigitalTechnologies #AI #ResearchPaper

Warm Congratulations to Ryt Bank  on Winning the PIKOM Digital Excellence Awards 2025 - Artificial Intelligence! 🎉 We are proud to have worked alongside Ryt Bank's journey in becoming the leading AI Bank, and are inspired by their commitment to pushing the boundaries of technology. We look forward to continuing our partnership and exploring new possibilities together to drive impactful advancements in AI for Malaysia. Once again, congratulations to the entire Ryt Bank team!

🚀

👏👏👏

🚀 We are proud to celebrate the successful conclusion of The Deepfake Detection, Localization and Interpretability workshop at IJCAI 2025, held in Montreal! 🧠The International Joint Conference on Artificial Intelligence (IJCAI) is one of the most prestigious top-tier conferences in the AI field, covering a broad range of research areas. 🧑🏫This workshop showcased groundbreaking advancements in deepfake detection, localization, and interpretability, while fostering global collaboration to tackle these challenges. 💽 Ant Digital Technologies also officially open-sourced the 1.8 million-sample DDL-Datasets, covering face forgery, video manipulation, and voice cloning with over 80 forgery techniques. With precise annotations of manipulated regions and timestamps, this dataset aims to enhance algorithm interpretability and drive innovation in the fight against deepfake threats. ️ 🎙️The keynote session highlighted advancements in deepfake research: Prof. Anderson Rocha on disinformation detection, Prof. Zhen Lei on wavelet analysis and joint learning for improved accuracy, and Prof. Abhinav Dhall on multimodal detection across cultures and languages. 🏆 The presentation session highlighted some of the brightest rising stars. Human Action CLIPs earned the Best Paper award, while Synlab, the Best Paper Candidate featuring collaborative visual misinformation detection in videos, brought together innovative ideas from four leading universities. Additionally, Xi’an Jiaotong University demonstrated their advanced multi-modal deepfake detection using the FPN-Transformer. 🎉 Also, congratulations to the winning teams, including FaceVolume, Stellarwave, and TrackKiller, for their exceptional contributions! ️ ❤️A heartfelt thank you to our outstanding facilitators Jianshu Li and Joey Tianyi Zhou for their leadership in making this event a success! We've also been encouraged by the positive feedback received from attendees after the event 🌐Together, we can outsmart deepfake technologies and build a safer digital world. #AIforGood #DeepfakeDetection #IJCAI2025