RPI ID: 2022-058-301, 2022-058-401
Innovation Summary: This invention introduces a self-supervised representation learning (SSRL) system that utilizes multi-segmental informational coding to enhance data representation. The SSRL circuitry includes transformer-based components designed to process input data without requiring labeled examples. By leveraging multi-segmental encoding, the system captures richer contextual relationships across data segments. This approach improves the quality and robustness of learned representations in machine learning tasks.
Challenges / Opportunities: Traditional supervised learning methods depend heavily on labeled datasets, which are costly and time-consuming to produce. This technology addresses the challenge by enabling effective learning from unlabeled data. It opens opportunities for scalable AI applications in domains where labeled data is scarce. Additionally, it enhances model generalization and performance across diverse data types.
Key Benefits / Advantages: ✔ Eliminates need for labeled training data ✔ Enhances contextual understanding through multi-segmental encoding ✔ Improves model robustness and generalization ✔ Reduces training cost and time ✔ Compatible with transformer architectures ✔ Scalable across various data modalities
Applications: • Machine learning and AI systems for image, text, and signal processing
Keywords: Self-supervised learning, representation learning, transformer, multi-segmental encoding, unlabeled data, AI
Intellectual Property: Patent Application No.WO2023244567A1 published, Patent Application No.18/874658 pending