August 2024 Summaries

The global AI market is expected to grow exponentially, with a projected value of $196.63 billion by 2024 and a CAGR of 28.46% between 2024 and 2030. The computer vision market was worth $20.31 billion in 2023, with a projected CAGR of 27.3% between 2023 and 2032. Natural language processing (NLP) is expected to reach $31.76 billion by the end of 2024, growing at a CAGR of 23.97%. Over 50% of US companies with more than 5,000 employees use AI, while Chinese and Indian companies report the highest use of AI compared to other developed countries. The large language model (LLM) market is currently valued at $6.4 billion and is expected to reach $36.1 billion by 2030, growing at a CAGR of 33.2%. AI adoption is widespread, with over 34% of companies already adopting AI, while 22% plan to do so by the end of 2024. The insurance industry has the highest AI adoption rate, followed by US healthcare companies. Companies that lead in AI functionalities produce total shareholder returns (TSR) four to six times higher than organizations that lag in AI investments. AI is expected to contribute around $15.7 trillion to the global economy by 2030, more than India and China's current GDP.

Multimodal datasets are like the digital equivalent of our senses, combining various data formats such as text, images, audio, and video to offer a richer understanding of content. These datasets allow AI to catch subtleties and context that would be lost if it were limited to a single type of data, providing advantages in tasks like image captioning, sentiment analysis, medical diagnostics, and more. Multimodal deep learning involves using deep learning techniques to analyze and integrate data from multiple sources simultaneously, enhancing model performance in various applications. By combining visual data with other modalities and data sources, models can achieve higher accuracy in tasks such as object detection and image segmentation. Multimodal datasets also allow models to learn deeper semantic relationships between objects and their context, enabling more sophisticated tasks like visual question answering and image generation. These datasets are crucial for advancing research in computer vision, large language models, augmented reality, robotics, text-to-image generation, VQA, NLP, and medical image analysis. By integrating information from data sources of different modalities, models can better understand the context of visual data, leading to more intelligent and human-like large language models.

Modern AI is moving beyond traditional machine learning models, requiring more sophisticated frameworks that can perform complex inferences on extensive datasets. However, as model complexity increases, so does the need for interoperability among multiple frameworks used to build, test, and deploy AI systems. The Open Neural Network Exchange (ONNX) framework addresses this challenge by offering a standardized, open-source format for representing AI models, allowing developers to seamlessly integrate AI tools with existing tech stacks. ONNX provides key features such as open-source support, standardized format, conversion tools, visualization and optimization libraries, interoperability, focus on inference, format flexibility, performance optimizations, and compatibility with popular frameworks like PyTorch, TensorFlow, Scikit-Learn, Keras, Microsoft Cognitive Toolkit (CNTK). It also offers pre-built conversion libraries to simplify the process of converting models from various frameworks to ONNX. With ONNX, developers can build, share, and run models across multiple platforms without worrying about compatibility issues, thereby streamlining the entire model development and deployment lifecycle.

Encord has raised $30 million in Series B funding to further invest in its multimodal AI data development platform. The company aims to be the final AI data platform a company ever needs, having already assisted over 200 top AI teams with strengthening their data infrastructure. Encord's focus on creating high-quality AI data for training and validation has led to better model performance for its customers. With the new funding, Encord plans to accelerate its product roadmap, continue innovating in the data layer, and launch a new end-to-end data management platform called Encord Index, which will make it easier for users to manage and curate multimodal data at scale. This platform aims to bring ease to data management and governance, enabling AI teams to understand and operationalize large private datasets in a collaborative and secure way.

The future of video annotation is here, thanks to Meta's Segment Anything Model 2 (SAM 2), a new foundation model that extends the capabilities of the original Segment Anything Model into the video domain. SAM 2 integrates advanced segmentation and tracking functionalities within a single, efficient framework, enabling real-time object tracking, memory module, improved performance, and efficiency. The SA-V dataset, created using the SAM 2 data engine, is an extensive collection of video annotations designed to support the development and evaluation of advanced video segmentation models. With SAM 2's interactive model-in-the-loop setup, annotators can refine and correct mask predictions dynamically, significantly speeding up the annotation process while maintaining accuracy. The SAM 2 data engine addresses the challenge of starting from scratch by progressively building up a high-quality dataset and improving annotation efficiency over time. SAM 2 has been shown to be faster, more efficient, and maintain quality with each phase of its development, and is now available for use in Encord's automated labeling suite.