Authors: Meena Jagadeesan, Michael I. Jordan, Jacob Steinhardt, Nika Haghtalab

Published on: June 26, 2023

Impact Score: 8.15

Arxiv code: Arxiv:2306.14670

Summary

• What is new: Shows competition can alter scaling trends in machine learning, potentially decreasing overall predictive accuracy instead of improving it.
• Why this is important: Scaling laws suggest increasing model size improves accuracy, but this doesn’t consider competition among model providers.
• What the research proposes: A new model of competition for classification tasks that uses data representations to study the impact of scale increases.
• Results: In many cases, improving data representation quality decreases overall predictive accuracy across users in marketplaces with competing model-providers.

Technical Details

Technological frameworks used: nan

Models used: Closed-form formulas, simulations with pretrained representations on CIFAR-10

Data used: CIFAR-10

Potential Impact

AI and machine learning marketplace, particularly those involving classification tasks and competing model providers.

Want to implement this idea in a business?

We have generated a startup concept here: OptiScale AI.