Optimizer Choice Impacts AI Forgetting, UC San Diego Tool Teaches 25 Million to Code

A recent study reveals that the choice of optimizer—such as SGD, RMSProp, or Adam—significantly impacts catastrophic forgetting in machine learning models, playing a larger role than previously understood (via an academic paper).

According to a UC San Diego announcement, a coding tool teaching 25 million people to code is even more critical in the age of AI, as understanding code remains essential because AI can make mistakes.

Accrete, Inc., a dual-use AI company founded in 2017, announced that Lieutenant General (Ret.) Daniel L. Karbler has joined as a strategic advisor (per a company press release).

Gymea High School is seeing success with its new mandatory in-house AI learning tool, as teachers are embracing AI to guide students and ensure honesty (according to local news).

Research outlines three ways to measure catastrophic forgetting in AI: retention (performance drop), relearning (reacquisition speed), and activation overlap (shared internal representations), providing insights into how AI systems forget (via an academic paper).

A post on TechAnnouncer explores the nuances between "automatization" and "automation" in modern technology.

Sources:

• St George News

• Hackernoon

• Hackernoon

• Techannouncer

• Finger Lakes Times

• Uc San Diego Today