Source: Regularized Deep Learning Memes

Accelerating Artificial Intelligence Initiatives

There are 2 places in the world that has most wait time

1. Airport where people are transiting or waiting for their loved ones
2. Organizations where Data Scientist are waiting for their model training to complete

Data scientist typically have to go through 100’s of training iteration for each hypothesis to narrow down on feature and model selection, hyperparameter configuration among others. Data in many cases are way beyond what a single node or GPU can fit in. For large datasets each iteration can go upward of days.

Enterprises hiring artificial intelligence and machine learning expert without right infrastructure and tools is like

Hiring astronauts to drive a bullock cart

Building data science capability within enterprise must be thought ground up right from selection of silicon chip.

Below is bare minimum necessity for AI driven organization to accelerate cycle of Hypothesis to production

• Infrastructure with right kind of hardware (GPU, CPU, HPC etc), technologies (Hadoop, Kubernetes etc.) and tools (Spark ML, Tensorflow, scikit etc.) that can distribute ML/DL pipelines
• Centralize data, build key pre-engineered features and provide functionality for feature sharing across use cases

Be Data First before thinking of being AI First

• Improve agility with self-service access to data and tools that enable seamless data exploration and data preparation
• Capabilities that can rapidly operationalize models, automate data engineering (to extent possible), monitor models and Identify data drifts

Originally published on my linkedin — https://www.linkedin.com/pulse/accelerating-artificial-intelligence-initiatives-srivatsan-srinivasan/