Our services in the selected country:
  • No services available for your region.
Select Country:
Remember my selection
Your browser is out of date. It has known security flaws and may not display all features of this and other websites

GSAM Perspectives

Machine Learning Decoded


Note: Separate multiple email address with a comma or semicolon.

Send me a copy


Note: Separate multiple email address with a comma or semicolon.

Your Name:

Your Email Address:

Send me a copy

The ability to process data, gleaning insights that were previously hidden from plain view, and converting them to actionable information is critical to success for data-driven investors

Today’s analysts have access to a never ending stream of data, outpacing even the most prolific experts. By any metric, we are creating more data than ever before and at an ever-increasing rate. Simply having faster computers and bigger databases does not, by itself, solve the predicament of digesting this massive quantity of data. As a result, data processing algorithms have evolved from simply processing to learning how to process. This approach is called machine learning.

Machine Learning

Machine learning has roots in artificial intelligence. It creates relationships between known data points and uses those relationships to make predictions on new data. The past two decades have seen advances in theoretical and practical machine learning technology, and an increasing need for accurate machine learning algorithms to tackle the unprecedented amounts of data today.

There are two basic types of machine learning algorithms: supervised and unsupervised. Supervised machine learning algorithms are used to make predictions based on historical observations. These algorithms analyze historical data (called training data) and model the relationship between input data (defined by its ‘features’ in machine learning parlance) and labeled output data. Handwriting recognition is one innovative application of supervised machine learning, where a supervised machine learning algorithm looks at a large set of pictures of handwritten alphabets. These pictures are generally pre-labeled with the actual letter they contain. The machine learning algorithm therefore learns the relationship between the input data (pictures of letters, pixel colors and intensity) and the output data (the actual letter). A successfully trained supervised machine learning algorithm can then be used to make predictions on live data. The US Postal Service, for example, uses this technology to read handwritten letters and sort them by address.1

The goal of unsupervised machine learning is to analyze a large set of input data in order to create structure around it. One application of this process is to analyze data with the objective of classifying it into particular categories. For example, unsupervised machine learning can be used by apparel makers to decide how to size their small, medium and large-sized t-shirts. This process would begin with data on the target customer population, such as height, weight, etc. An unsupervised machine learning algorithm could then analyze the distribution of heights and weights in the sample and create three segments that contain a population set of similar height and weight characteristics. It can also define a natural delineation between the other segments and determine how the dimensions of a particular t-shirt differ from other sizes.

There is a finite group of algorithms that can aid in supervised or unsupervised machine learning. The choice of which algorithm is best suited for a particular application depends on the data being analyzed and the purpose of the analysis. Ultimately, successful machine learning algorithms allow us to create intelligent processes that can generate predictions -- an invaluable asset in today’s data-driven world.

Self-Driving Cars

In the US, the average person spends over 50 minutes a day commuting to work,2 of which 20% is typically spent in traffic, representing approximately $42 billion a year in wasted fuel.3 In addition to the costs of congestion, the National Highway Traffic Safety Administration (NHTSA) estimates that traffic accidents cost almost $826 billion annually in the US.

Self-driving cars (also known as autonomous vehicles) have the potential to reduce commute times, allow drivers to utilize their commute more efficiently and over time improve the safety of driving. This means avoiding many of the preventable, human causes of crashes, including fatigue, texting, alcohol or speeding. A self-driving car will not get distracted and has been programmed to constantly develop a back-up plan in case of any identified threats. For every vehicle on the road, self-driving cars can generate a defensive plan based on a multitude of potential situations. If an adjacent car happens to swerve or come to a sudden stop, the driverless car’s computer algorithms can react within milliseconds. If it works as intended, autonomous driving has the potential to greatly improve road safety. However, as with any new technology, it will come with its own unique safety concerns. How the automobile industry continues to navigate these safety issues as they arise will be critical to the technology’s success.

Self-driving cars’ ability to drive autonomously relies heavily on a similar learning process to humans. They utilize a suite of 360-degree cameras along with a spinning laser that enables them to measure the exact distances to objects in their vicinity. All this data is digested through sophisticated machine learning computers that interpret the surrounding environment much like a human would do while driving, in real-time. The computer algorithms combine pre-defined driving rules (i.e. stop at a red light) with real-life empirical evidence gathered through actual driving experience, in an effort to become better drivers. Human drivers, in a similar fashion, typically learn through examination of pre-defined driving rules, supplemented with behind-the-wheel driver training. All of this works in concert to create a more comfortable, productive and safer commute, resulting in truly powerful benefits.

Precision Agriculture

In a practice known as “precision agriculture,” farmers use big data to improve their productivity and the quality of their crops. While best known for its seed and pesticide businesses, some large agricultural companies have been willing to make large investments that enable them to make decisions based on big data.

By analyzing real-time field measurements collected by drones and remote sensors, climate data and the growth characteristics of thousands of crops, farmers are able to determine the optimal day to plant certain crops. This detailed analysis, combined with information about the topography of certain fields, can also establish the ideal depths and spacings for millions of individual seeds. Using a planter loaded with this type of data and equipped with a GPS, farmers can more precisely plant their fields. Farmers who utilized this technology-driven system reported crop yields that increased by almost 5% over two years, a more significant improvement than any other recent innovation.4


Machine learning is transformative as it can help make predictions using complex datasets in almost any environment—ultimately gleaning powerful insights that were previously hidden from plain view. Having more data, in itself (and there is certainly much more data today than ever before), does not necessarily empower companies or people to become more efficient. Simply having bigger and faster computers does not necessarily lead to intuitive outcomes in a real-life, human environment. Machine learning allows computers to learn very much like a human would learn—by operating with pre-defined rules of the road (i.e., a driver’s manual), but then more importantly, learning from empirical evidence as more data is gathered and interpreted (i.e., learning to drive by actually driving). Automation and gains in productivity have been at the core of every technological revolution (agricultural, industrial, computer and data revolutions) over the past two centuries—and machine learning advances these gains into new territory, potentially driving significant results. Nonetheless, the necessity of human oversight and intuition cannot be understated. A computer can beat a human at chess, but we believe a computer with a human can beat any computer by itself.

Related Perspectives

GSAM Perspectives
Text, Tone and Topic

While computers today may not yet be as adept with language as humans, natural language processing will become a crucial part of the investing process in the coming years.

GSAM Perspectives
The Role of Big Data in Investing

GSAM’s Quantitative Investment Solutions team shares their synthesized views on the role of big data in potential investments and key techniques for approaching big data.

GSAM Perspectives
Big Data is Fundamental

GSAM’s Fundamental Equity and Fixed Income teams discuss how big data can be a strategic advantage or disruptive force in companies and sectors.