Sensational Machine Learning Services

What Is Machine Learning?

Machine learning (ML) is the process of teaching algorithms to make decisions, based upon data. It encompasses a range of sub-domains like reinforcement learning and plays a crucial role in the discipline of data science.

Organizations use ML primarily to automate decisions that humans would otherwise perform. The major benefit of outsourcing decisions to ML is that it is quantifiable and enforced.

Humans are free to make their own decisions, which is both a blessing and a curse. It is problematic when different people make decisions based upon their own rules.

Take an application form, for example. If humans applied their own judgement then two different people would make two different application decisions. ML encodes the knowledge of all humans to make an unbiased, systematic, traceable assessment.

The downside of ML, however, is that it is hard to encode exceptions. Humans are quick to react to exceptional circumstances, but ML often struggles because it hasn’t observed enough of these states. In these cases human-in-the-loop approaches are beneficial.

What is Machine Learning Not?

Machine Learning is not Artificial Intelligence

Artificial intelligence (AI) is an academic philosophy debating human consciousness. When does a programmed machine become human-like?

Machine learning (ML) is the specific process of teaching an algorithm based upon data. It can only learn from the data you provide it. It does not “think”.

ML in industry is generally considered to be an extension of an “expert system”, which is an application that is specifically designed to perform one task with super-human precision.

This means that ML solutions are targeted. They are designed by experts to automate a specific problem.

ML is not A Silver Bullet

ML learns to make a decision based upon the data it can observe.

This means to make ML work, you need the data to learn from. Often we can work with a small amount of data, or create simulations to generate more data, but models need a decent amount of data to make them robust.

ML is applicable to a large number of problems and situations, but not all. As experts in the industrial application of machine learning, we can help you find problems that are a good fit for ML. If you’re unsure then send us a message and we’de be happy to help.

How Does ML Help?

Machine learning (ML) helps businesses automate decision making processes.

As well as optimizing current business operations, ML can also lead to new products and features. For example, the finance industry has developed a wide variety of ML models to automate routine tasks (e.g. FinTech). Whereas the Technology industry enables whole new industries through novel applications of ML (e.g. Google).

ML automates single-shot, point-in-time decisions and forecasts. Single-shot means that the algorithm is trained to make correct predictions for that single observation only. For example, predicting that a picture of a t-shirt is of a class named t-shirt, or predicting the cost of some component in 6 months time.

ML does not optimize over multiple steps. For example, if you’re attempting to predict what product to show next on a website, then that item depends on what the user has previously browsed. If you used ML in this case, the ML algorithm would suggest showing the same second item, no matter what the user browsed first. Reinforcement learning is used in these cases.

In our experience developing ML solutions for organizations like Google, Microsoft, and Shell, given the right situation, there are a number of benefits ML can provide:

• Automation of decision intensive tasks: We’ve worked in the distributed acoustic sensing (DAS) market for a long time. These are systems that produce vast quantities of audio data. You can classify signals by eye, but there is just too much data to do that consistently all the time. ML automates this process to produce activity detection classifiers. This idea can be applied to all other industries and domains.
• Consistent, quantitative decision making: One key problem when scaling a business is that different people make different decisions. Using ML to make a decision allows you to make decisions more consistently, faster. It’s also hard to quantify human errors and biases, whereas this is standard practice within ML.
• New product lines: Many companies exist because outsourcing a business or user function to a service is easier than trying to do it themselves. If you spot a situation where you own or can collect a large amount of data about a task, it’s often possible to leverage that to offer new products or services. The key, however, is finding a problem that is worth solving. This is the principal reason why ML projects “fail”. Solving a problem that doesn’t need solving.
• Reducing operational burden: We’ve work on some projects that aim to reduce the operational burden of a business. For example, we worked with Shell to build a domain-specific question and answering tool to answer questions like “what valve fits onto a T-2304”. If you can automate common tasks or decisions, this gives free time back to your staff, to concentrate on more important tasks like product development or sales. The key with this type of work is to find ways to make their life easier, not attempt to automate them away.