Term
Zero-shot learning (ˈzɪər-əʊ ʃɒt ˈlɜːrnɪŋ)
Definition
Zero-shot learning is an AI method that allows a system to correctly handle tasks it hasn't been specifically trained to perform. It uses prior knowledge to make intelligent guesses about new tasks.
Where you’ll find it
Zero-shot learning is often featured in AI development platforms, particularly in sections related to model training and capabilities. It is generally available in more advanced AI plans or tools that emphasize adaptable, generalized AI models.
Common use cases
- Allowing AI models to classify objects or concepts that were not included in their training data.
- Increasing the adaptability of AI in applications where it can encounter unpredictable scenarios, such as robotics or customer service bots.
- Reducing the need for extensive dataset compilation by leveraging the model's ability to intuit new information.
Things to watch out for
- Zero-shot learning may not always achieve the accuracy of models trained on specific tasks, particularly in complex scenarios.
- It may require sophisticated configuration or tuning to work effectively across different types of tasks.
- Understanding the limitations of this approach in practical applications is important to avoid overestimating the capabilities of the AI.
Related terms
- Transfer learning
- Generalization
- Supervised learning
- Unsupervised learning
- Machine learning