Machine Learning

Machine learning is the idea that there are generic algorithms that can tell you something interesting about a set of data without you having to write any custom code specific to the problem.  Instead of writing code, you feed data to the generic algorithm and it builds its own logic based on the data.  It's really just an application of artificial intelligence algorithms that gives a computer access to large amounts of data and enables it to figure out solutions on its own.

Machine learning model - decision tree model

Classification algorithm

Cost function

neural networks

fuzzy logic

object recognition

confidence

Focus on problems where a human could solve the problem, but where it would be great if a computer could solve it much more quickly.

 

To mitigate the risks associated with the erroneous behaviors of machine learning systems and provide users with the best possible experiences, designers should take the following steps before releas‐ ing their software to the public:

  1. Design for interactions in which the system explicitly restates its understanding of the tasks it has been asked to perform, giving users the chance to catch errors and redirect the system’s behav‐ ior.

  2. When possible, provide fallback mechanisms through conven‐ tional user interfaces that allow users to circumvent machine- learning-enhanced functionality and perform tasks using explicit logic and interfaces.

  3. Perform rigorous testing of the software in as many environ‐ ments and usage scenarios as possible to uncover possible faults or inconsistencies that may arise from conditions that differ from those of the original development environment. Limited release to an audience of more knowledgeable and error-

    tolerant testers may help to uncover circumstantial inconsisten‐ cies before wider release to the public.

    1. Use all available metrics, namely confidence scores, to assess the validity of included features. Set realistic expectations in how you present a feature and its effectiveness to the user.

    2. Resist the temptation to include an impressive-sounding feature if its behavior is too unreliable. This assessment should weigh the complexity and potential value of the feature against the likelihood of its failure. If a feature offers some potentially revo‐ lutionary new capability, users may be more willing to accept that it only works 90% of the time.

    3. Make users aware of any risks that might accompany their use of the software or a particular feature and allow them to decide for themselves whether these risks are outweighed by the poten‐ tial benefits of the system’s functionality.

    4. In cases where a system failure may have extraordinarily serious consequences such as irrevocable damage to the user’s property, injury, or death, the value of a feature’s inclusion should be weighed with extreme caution and a lawyer should be consulted to assess the risks and liabilities as well as to formulate any nec‐ essary disclaimers that will be presented to users.

    5. Most machine learning challenges relate to handling your data and finding the right model.

      Data comes in all shapes and sizes. Real-world datasets can be messy, incomplete, and in a variety of formats. You might just have simple numeric data. But sometimes you’re combining several different data types, such as sensor signals, text, and streaming images from a camera.

      Preprocessing your data might require specialized knowledge and tools. For example, to select features to train an object detection algorithm requires specialized knowledge of image processing. Different types of data require different approaches to preprocessing.

      It takes time to find the best model to fit the data. Choosing the right model is a balancing act. Highly flexible models tend to overfit data by modeling minor variations that could be noise. On the other hand, simple models may assume too much. There are always tradeoffs between model speed, accuracy, and complexity.

      Every machine learning workflow begins with three questions:

      • What kind of data are you working with?

      • What insights do you want to get from it?

      • How and where will those insights be applied?