In a professional kitchen, everything has its place. Everyone has their role to play. Entry-level prep cooks mise the ingredients, dicing onion after onion and peeling potatoes until they can no longer feel their fingers. Other cooks work different sections of the line, from salads and sauces to grilling meat and plating the final dish. As each dish is prepared, other chefs taste, season, and adjust along the way until the best possible dish makes it to the hungry diner. In many ways, a professional kitchen is a lot like deep learning.
Deep learning is a more advanced level of machine learning, which itself is an advanced level of artificial intelligence (AI). If AI is a home cook preparing dinner with the tools and cookbooks available to them, machine learning is a trained chef who uses their knowledge and skillset to improve upon old recipes and create new ones from scratch. The chef fine-tunes the basics. Deep learning, however, is far more advanced — it is a fully-staffed professional kitchen with each layer of the dish belonging to a different chef, with that dish getting constantly tasted and adjusted, with each chef contributing to a final dish that is more complex than what any one chef could create alone, with the kitchen taking in feedback from diners and refining … and refining and refining.
In literal terms, deep learning is a layered structure of an AI model, with each subsequent layer able to focus on different and more specific parameters.1 It is designed to mimic the neural networks in the human brain and can self-correct and improve over time, establishing the connections among and between the layers itself.1,2 Deep learning is more autonomous and accurate than machine learning and requires less human intervention.2
Deep learning models can be further refined when trained on vast quantities of unstructured and unlabeled data,2 like a kitchen full of chefs making food based on their learned skills, cultural influences, professional experiences, sensory cues, verbal feedback, and even the restaurant’s budget. Such fine-tuned deep learning foundation models are more nuanced and include LLMs like ChatGPT.
Before ChatPT evolved into a deep learning model, it had to be built on tons of data that humans heavily cleaned, labeled, reviewed, and classified. Deep learning itself has improved over time. From 2015 to 2021, the cost to train an image classification system fell 64% and the time it took to train deep learning models improved by 94%.2 But deep learning powers far more than just chatbots. It has already helped detect skin cancer, aided drug-discovery, improved the efficiency and sustainability of farming and modernized manufacturing.
TrueShares Technology, AI & Deep Learning ETF (LRNZ) is built to capture the diverse layers of AI, including tech companies employing and/or enabling advanced applications of deep learning in a variety of sectors. LRNZ generally holds 20-30 positions we think are poised to become category killers. It seeks the next up-and-comer in tech like a foodie seeks the next hot restaurant.