Artificial intelligence … and all that jazz : Advice for Students (10/n)

Writing this out of turn in my series. Was called to give a keynote in a conference, so thought of taking down my notes and writing this for the benefits of engineering students and academicians especially in computer science and electronics streams. The AI became more virulent in the last 3–5 years and is touted as the next in thing. Either you are doing AI or you are a dumb engineer or an organization. While AI surely is a forward looking. Solves multiple use-cases and would be more visible in almost all the tech products in the coming years. At a more sublime level it gives a guilt-free framework for the system to make decision- right, wrong or partially correct.

To implement an AI system following key skills are required. The students and academicians can judge themselves depending on their preference, which path to master and specialize. It’s always a good idea to be aware or proficient in all the departments. As with all the sweeping generalizations there is a good possibility of missing the finer point. Caution must be exercised in forming any opinion.

1. Domain Knowledge

This is a classical path for engineers, by far the most important career path. The core industry sectors are still going to remain e.g Automotive, Industrial, Smart utilities, Avionics, Power, Space, telecom, IT .. so on an so forth. Bulk of the jobs would still be attributed to the core engineering. Just because of the advent of AI, these sectors won’t just vanish, they’ll be benefitted from it. Therefore my advice would be to focus on the core engineering and keep your eyes and ears open for AI remedies specific to your sector. In your new job down the line you’ll be integrating or generating data for the industry specific AI model.

2. Data Scientist

Do you need an engineering degree to be a data scientist? Probably not, however engineers do have an advantage due to the mathematical rigour they have to undergo especially signal processing, electromagnetics or optics. I would say statisticians or mathematicians with a good hand on programming are best suited for this role.

Data scientists are the guys who know how dress the data to apply advanced models and simulate the desired outcomes.

Typically in an organization, a very small team(relatively speaking) under an expert is sufficient to design data, data models and simulate various recipes. Once the blueprint is ready, the engineers take it forward to scale and port on various platforms. This is where the bulk of jobs are, we”ll talk about it in next sections.

3. Frameworks: AI modelling and simulation

Here’s a verbatim quote from NVIDIA : “Deep learning frameworks offer building blocks for designing, training and validating deep neural networks, through a high level programming interface. Widely used deep learning frameworks such as MXNet, PyTorch, TensorFlow and others rely on GPU-accelerated libraries such as cuDNN, NCCL and DALI to deliver high-performance multi-GPU accelerated training”

Let’s dissect it a bit and see where your preference lies

1. Deep learning frameworks offer building blocks for designing, training and validating deep neural networks, through a high level programming interface.

Data scientists along with the help of engineers would be expected to program the data and implement models in the high level languages. Closest analogy is Java programmers for implementing a particular module. The key skill here is the knowledge and proficiency of using the framework. Programming skills are must BUT engineering degree …not necessary. (assumption: Engineering degree is not equal to learning a programming language).

2. Widely used deep learning frameworks such as MXNet, PyTorch, TensorFlow and others.

There are many frameworks and considerable work is being done in each one of them. This is where the bulk of jobs would come. Therefore students should focus on getting comfortable and writing code equally well in atleast 2–3 frameworks to maximize the possibility of employment.

3. …. rely on GPU-accelerated libraries such as cuDNN, NCCL and DALI to deliver high-performance multi-GPU accelerated training.

This is where I would invest my money on! To be able to efficiently use GPU or underlying hardware for running the models you need a very solid understanding of both the GPU or computing hardware architecture and an excellent software skills to use it effectively. This requires computer engineers to be familiar with electronics and Electronics engineers to be proficient in writing software and all of them being aware of cloud computing platforms as well[ aws, azure…].

This is a painful path and won’t bring the instant glory or money but this kind of breed is very well sought after in the industry, worst case you’ll end up becoming an excellent framework programmer.

4. Scaling on cloud and GPUs

Deep learning algorithms and model trainings typically use cloud or on prem server resources which costs real dollars. This is the most crucial step in the deployment of any AI strategy in an organization. Therefore the real skills required are to implement efficiently the models developed by the data scientists to utilize the power of state of the art hardware using least amount of dollars. This would come after few years of experience in handling GPU, Cloud and frameworks.

5. Scaling on Embedded devices

With the proliferation of IoT devices, most of them battery operated. It would not be practical to bring all the data and compute to the cloud. The new generation of IoT devices would be capable of implementing machine learning on the edge. This is implementing the AI algorithms on constrained environments. This would be the most sought after skill in the coming times. Embedded engineers who have an eye on AI are best suitable for this job.

To summarize the AI skills would dovetail into the existing engineering streams, whereas the programming part on frameworks and data science is up for grabs by people from diverse backgrounds not necessarily engineering.