Page 149 - Computer - 7

Limitations of Machine Learning Systems

AI systems are subject to failures resulting both from accidents (safety failures) and from adversarial
malicious activity (security failures). There are many different types of Machine Learning failure modes,
but perhaps the most common is when the training data is not sufficiently representative and instructive
for the diverse, real-world examples the Machine Learning system will encounter.

For example, a satellite imagery classifier that is trained to recognise vehicles exclusively using training
data images in a desert environment should be assumed to have degraded performance if the operational
data images are of the same vehicles in a grassland, urban, or arctic tundra environment. For the same
reason, the performance of ML models in real world applications generally degrades over time if not
regularly updated with new training data that reflects the changing state of the world.

Database Subject Enrichment

Image recognition is one of the most common applications of Machine
Learning. It is used to identify objects, persons, places, digital images, etc.
The popular use of image recognition and face detection is Automatic friend
tagging suggestion used by social media apps. Facebook provides a feature
of auto friend tagging suggestion. Whenever we upload a photo with our
Facebook friends, we automatically get a tagging suggestion with name,
and the technology behind this is machine learning’s face detection and
recognition algorithm. It is based on the Facebook project named Deep Face, which is responsible for
face recognition and person identification in the picture.

Post-Processing  


 The definition of Artificial Intelligence changes with the discipline of study to which it is applied.

 The various disciplines of study that approach AI are Computer Science, Computer Engineering,
Philosophy, Psychology, Mathematics & Statistics, Neuroscience, Linguistics, and Biology.
 Symbolic AI refers to approaches to developing intelligent machines by encoding the knowledge
and experience of experts into sets of rules that can be executed by the machine.

 Symbolic AI uses Expert Systems or Fuzzy Logic Systems as the underlying algorithms.

 Machine Learning AI systems generate their own rules without human intervention by observing
and recognizing patterns.
 Factors that have led to growth in focus on Machine Learning include massive datasets, increased
computing power, improved algorithms, and open-source code libraries and frameworks.

 Machine Learning can be Supervised, Unsupervised, Semi-supervised, and Reinforcment.

147