ML on the edge

Excited to learn about the new Visual lookup in Apple Iphone which recognizes plants and trees, seems like a fascinating use case for "ML on the edge". Here is the more detailed steps on how to do this from apple.

Although users you can claim this feature has already been available through purpose built apps like “picture this and seek for identifying plants, Merlin for identifying birds species" and also the mighty google lens in android platforms which can identify locations etc. ?

It shows embracing AI or embedding AI functionality has become imperative for Product organizations to boost their competitive edge and user experience.

As part of my MLOps training, I am excited to learn about the following 2 tuning techniques. Also able to relate that these techniques can help to reduce the size of bigger models enabling them to run on edge devices.

• Pruning : Elevating Efficiency! Pruning takes center stage by trimming the excess in our language models. It's like sculpting for efficiency, removing redundant parameters and paving the way for leaner, meaner models. Less is more when it comes to optimal performance!
• Quantization : Scaling Down without Sacrificing Quality! Quantization steps in by reducing the precision of numerical representations. Think of it as finding the sweet spot between model size and accuracy. Smaller bit sizes, big impact - making models more accessible without compromising on quality.

Why Does it Matter?

• Resource Optimization: Pruning and Quantization make models more resource-friendly, ideal for various deployment scenarios.
• Scalability: Streamlining models ensures scalability without sacrificing performance, opening doors for broader applications.

Read about recent developments on pruning techniques from Andrew Ng's recent weekly newsletter - "The Batch".

Further research developments in these areas are fascinating tends to improve adoption of LLMs/Image models on edge devices.

Further Reading / References :

• SparseGPT: Massive Language Models Can be Accurately Pruned in One-Shot
• A Simple and Effective Pruning Approach for Large Language Models
• Andrew Ng's weekly AI Newsletter - "The Batch"

Stay Curious and Happy learning!

#AI #MachineLearning #Pruning #Quantization