Zero-shot and few-shot learning — Deep Learning & Neural Networks

Traditional supervised learning requires hundreds or thousands of labelled examples per class. Zero-shot learning makes predictions on completely unseen classes without any task-specific training data. Few-shot learning makes predictions from just 1–10 examples per class. These capabilities emerged in large language models as an unexpected consequence of scale — GPT-3's 175B parameter model demonstrated that sufficiently large models could perform new tasks from natural language descriptions alone, without any gradient update. Zero-shot and few-shot learning are transforming how AI systems are deployed: instead of collecting and labelling thousands of examples, a practitioner can describe the task in natural language and get reasonable predictions immediately. For sports analytics, this means classifying cricket shot types, detecting player mentions, or generating match summaries from a brief task description — without building a labelled dataset. Understanding zero-shot and few-shot mechanisms — from BERT's entailment-based zero-shot to GPT's in-context learning — is essential for leveraging the full capability of modern language models with minimal data collection.

Analogy🏏Cricket

🏏 Think of it like cricket: The Duckworth-Lewis-Stern (DLS) method, the ICC bowling economy metric, and the net run rate each measure team performance but optimise for completely different things — DLS cares about wickets-in-hand, economy cares about runs-per-over, NRR cares about tournament-wide margins. Using the wrong metric to evaluate a bowler is like using MSE for classification: technically computable but optimising the wrong thing entirely. Just as you would never rank a spinner by batting average, you should never use MSE when your model predicts probabilities. Just as focal DLS adjustments down-weight easy chases and amplify close finishes, focal loss down-weights easy examples and amplifies hard minority-class examples. The choice of scoring metric defines what excellence means — and so does the choice of loss function.