MLCamp: 2017 Expedition

Hello!

This marks the start of my preparation for the 2017 expedition of our Machine Learning Study Camp.

All trends point to the fact that machine learning and artificial intelligence are critical to the future — think self-driving cars, robotics & automation, smart assistants, deep learning & style transfer etc. — making it critical for more developers to invest in learning about the concepts, tools & technologies required to participate in this ecosystem.

Traditionally, machine learning has been viewed as the domain for experts in data sciences — requiring a solid grasp of statistical analysis, linear algebra and data analysis tools & techniques. The focus here has been on creating custom learning models for vertical domains or industries. Thus, developers could instrument their applications to generate data on user interactions — the data scientist could then use a subset of this data as a training to define suitable functions that could predict the app user’s behavior or preferences for future contexts.

More recently though, we are seeing the emergence of cloud-based Machine Learning As a Service solutions where the complexities of “how” learning is done is masked from the user, and instead exposed via APIs that allow developers to benefit from these pre-trained models by posting inputs (e..g, data for various features) and retrieving outputs (e.g., the predicted label for classification). This is particularly true for popular problems like image recognition, speech recognition, logo or face detection etc. where companies like Google and Facebook can leverage their non-trivial databases of user data & media to create models that have fairly good accuracy for generic use.

In between the two extremes — data scientists (building custom models) and app developers (using pre-trained models via APIs) — is an intermediate area where motivated developers can start bridging that gap by getting a better understanding of core concepts (sciences) and related tools (e.g., Python data analysis libs, TensorFlow). This is the goal for our ML Camp.

In the first iteration, we are looking to create our own trails to the summit based on using freely-available but professionally-crafted content for concepts and tools. Subsequent iterations may alter the checkpoint locations (or count), or identify new trails that provide faster (or slower) progress to the top to suit diverse participants.

Let’s get started.