Supervised learning is great — it's data collection that's broken

2017-04-02 · Source: Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

The article contends that current dissatisfaction with supervised learning is misplaced, asserting that labelled examples remain an effective means of specifying computational objectives. It highlights that the fundamental problem lies not with supervised learning itself, but with the inefficient and often tedious processes involved in collecting and reusing human knowledge for data labeling. The author argues that while Artificial General Intelligence is still distant, supervised learning provides a robust framework. Instead of waiting for unsupervised learning to provide a universal solution, the author advocates for a focused effort on improving data collection methodologies and enhancing the reuse of existing human-annotated data to address the core challenges in machine learning development.

Key takeaway

For Machine Learning Engineers focused on model performance, recognize that improving data collection and human knowledge reuse offers greater immediate returns than solely pursuing novel unsupervised methods. Prioritize investing in robust data labeling pipelines and strategies for leveraging existing annotated datasets to directly enhance supervised model efficacy and reduce development friction.

Key insights

Supervised learning is effective; data collection and human knowledge reuse are the real problems.

Principles

• Labelled examples effectively specify computation.
• Dissatisfaction with supervised learning is misplaced.
• Fix data collection, don't just await AGI.

In practice

• Improve data collection processes.
• Enhance reuse of human knowledge.
• Focus on data quality over new paradigms.

Topics

• Supervised Learning
• Data Collection
• Data Labeling
• Human Knowledge Reuse
• Machine Learning Development

Best for: AI Scientist, Machine Learning Engineer, Data Scientist

Related on AIssential

• The AI Bottleneck: High-Quality, Human-Powered Data — High-quality, human-aligned datasets are crucial for AI systems to solve real-world problems effectively.
• Data filtering methods for training language models — Label error detection methods improve model performance, particularly on small, noisy datasets, outperforming random data removal.
• Deep Active Re-Labeling: Toward Noise-Resilient Annotation Efficiency — Deep Active Re-Labeling improves active learning by strategically re-annotating noisy data to enhance efficiency and data quality.
• Data is the new coffee — Effective data annotation, like coffee tasting, requires structured processes, agreement, iteration, suitability, and cross-functional collaboration.
• Data Labeling Services for AI/ML: Building the Foundation of Intelligent Systems — High-quality labeled data is fundamental for training accurate and reliable AI and machine learning models.
• General Supervised Learning Framework for Open World Classification — A data- and model-agnostic framework automates unknown class identification, categorization, and model updates in open-world classification.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Explosion · Developer tools and consulting for AI, Machine Learning and NLP - Explosion.ai.