There are two broad categories of ML that have evolved: machines that “learn” using teaching examples or subsets, and machines that explore or discover facts and solutions on their own.Īn example of the first ML category might be an industrial robot that learns how to see if a component is properly assembled and located on a circuit board. The general concept has the words embedded themselves: methods for machines to “learn” on their own, without a priori writing of programs and algorithms. A classic early ML experiment involved a computer learning to balance a broomstick on an artificial hand ( Widrow, 1987). The advent of small, low-cost microprocessors in the 1970s allowed computerized implementation of prior analog AI, such as self-steering boat systems ( Francis West, 1979), following ES rules and algorithms to assert positive, intelligent control of many manufacturing, transportation, and related tasks.īy the mid-1980s, real-time minicomputers coupled with advancing analog-to-digital conversion chips allowed the development of early ML-based experiments and products. Mainframe computer programs in the 1960s were developed to automate manufacturing planning, based on organized bills of materials for components and subassemblies, and/or for early military attack and targeting. The earliest AI systems were analog mechanical devices, such as wind-vane self-steering for boats, centrifugal speed controls for engines, or automatic chokes and transmissions for automobiles. In addition, ES and ML tools can be combined into a single product or suite of products, such as the features offered by most contemporary antivirus software. In other cases, the human operator can take control if and when the computer can no longer make a successful interpretation of the data (e.g., a self-driving car might return control to the human driver if confused, overwhelmed, or specified safety parameters are exceeded, or a bank’s credit card supervisor might personally interview an applicant if the software cannot make a credit approval decision). An automated blood laboratory system, or breast cancer screening software, might be based on ML. For example, the computer can be taught or can infer patterns from a sufficiently large set of matching (correct) or defective (incorrect) examples that were previously scored or provided by one or more experts. The operator provides examples and positive reinforcement for correct decisions and examples and negative reinforcement for poor decisions.
In many cases, human operators also take part in training the ML system, and the process is much like training a pet. Cardiac arrhythmia detection algorithms used in automated electronic defibrillators (AEDs).īoth ES and ML system can be paired with one or more human operators to improve the quality, performance, or safety of the overall system.
0 kommentar(er)
