Ritu Jha-
As the Artificial Intelligence rush intensifies, developers have been scrambling to teach machines by labeling data for them. But in two emerging AI fields, machines also are teaching themselves by recognizing patterns in data.
It’s called Deep Topological Data Analysis, or DTDA, and Self-Supervised Learning, simplified as SSL, emerging fields that hold great promise across a wide variety of applications.
That was the message from Janhavi Giri, data scientist at Intel Foundry Manufacturing and Supply Chain. Giri spoke during the session, “Smart Manufacturing Session: Front-End Manufacturing – Optimizing Yield with AI at SEMICON West July 9-12 at the Moscone Center in San Francisco.
Giri Deep TDA is a type of artificial intelligence that helps us make sense of complicated data by focusing on its shape and how it’s organized. It’s like teaching computers to recognize patterns and connections in the data. Think of it as looking at a scattered bunch of dots and determining whether they form a circle, a spiral, or some completely different shape.
Self-supervised Learning is a way for AI to teach itself about data without needing many examples labeled by people. It’s like trying to solve a puzzle without knowing what the completed picture should look like; you have to figure out where each piece goes based on the shapes and patterns you see.
Giri said DTDA and SSL are helpful in different but complementary ways. Deep-TDA allows computers to understand the underlying structure and shape of data. It doesn’t just look at individual data points but examines how these points are arranged in relation to one another. This can be particularly useful when dealing with complex data where the relationship between points is more informative than the raw data itself.
Giri said SSL is advantageous because it reduces the need for large sets of labeled data, which are often expensive and time-consuming to create. Instead, it allows the AI to learn from the data itself by finding patterns and structures without explicit instructions. This is particularly helpful when there is a lot of data available, but it hasn’t been labeled by humans.
“Implementing these methods in semiconductor manufacturing can lead to significant advancements by automatically classifying and analyzing the images for rapidly identifying defects, predicting failures and optimizing manufacturing processes,” Giri said.
Giri said the two methods help to overcome the challenge of analyzing and extracting insights from complex data such as images in absence of their labels generated by humans. However, integrating these two approaches can lead to certain challenges such as computational complexity, integration issues, interpretability, generalization and scalability.
“Addressing these challenges requires a multidisciplinary effort, involving expertise in machine learning, data science, domain-specific knowledge, and computational resources,” Giri said.
By enabling more accurate and efficient analysis of image data, DTDA and SSL can provide significant benefits to any field where image segmentation and anomaly detection are critical. Giri said the potential applications are many:
Healthcare Industry – Medical professionals and researchers can use these AI techniques for analyzing medical images such as MRIs, CT scans, and X-rays to detect anomalies like tumors or fractures and segment different anatomical structures for diagnosis and treatment planning.
Manufacturing Sector – Quality control engineers can apply these methods to identify defects in products or components by analyzing images from production lines, leading to improved product quality and reduced waste.
Automotive Industry – Companies developing autonomous vehicles can use these techniques for real-time image analysis to identify and segment road features, obstacles, and anomalies, enhancing navigation and safety systems.
Agriculture – Agronomists and farmers can benefit from these AI approaches to analyze aerial images of crops for detecting diseases, pests, or nutrient deficiencies, enabling targeted interventions.
Security and Surveillance – Security professionals can use image segmentation and anomaly detection to monitor video feeds for unusual activities or objects, improving safety and response times.
Retail and Inventory Management – In retail, these methods can help in managing inventory by identifying and categorizing products through image analysis, as well as detecting anomalies in shelf arrangements.
Environmental Monitoring – Scientists and conservationists can employ these AI techniques to process satellite and drone imagery for tracking changes in ecosystems, detecting environmental hazards, or monitoring wildlife populations.
Research and Academia – Researchers in various fields can use Deep TDA and Self-Supervised Learning to analyze complex datasets, uncover new insights, and advance scientific knowledge.
Financial Services – For fraud detection, these methods can help analyze transaction patterns and identify out-of-the-ordinary behaviors that may indicate fraudulent activity.
