| |March 20199CIOReviewphones revolution driven by wireless connectivity and availability of cloud has allowed for virtually unlimited storage of data and unending abil-ity to churn it. However, there do exist few concerns about the impact of AI on our society and our future, the `technological singularity'. But as advancements and adoption of AI continue to accelerate, one thing is certain that the impact is going to be profound.Typically, machine learning and deep learning are tools/techniques under the umbrella of AI. To under-stand their application, let us look at self-driving cars.Today, machine learning al-gorithms are extensively used in self-driving cars. Integration of sensor data processing to an ECU (Electronic Control Unit) of a car allows the use of machine learn-ing to accomplish various tasks. Few applications include evaluat-ing driver's condition, lane keep assist, pedestrian identification and tracking among other objects of interest and analyzing the scenar-io. This is achieved by fusing data from different sensors like radars, cameras and/or the IoT (Internet of Things).Broadly, machine learning algo-rithms are two types - supervised and unsupervised. The difference is that the supervised learning requires training data to learn till they get to the level of confidence they as-pire for (the minimization of error). Supervised learning algorithms can be subcategorized into regression, classification and anomaly detection supported by dimension reduction.Unsupervised learning algorithms on other hand are more complex in the sense that, there is no training/ground truth data and their results require human (Subject Matter Experts') interpretation.Deep learning (DL) is a part of an extended family of machine learn-ing techniques based on training data representations as opposed to task-specific algorithms. Deep learn-ing architectures are recurrent neural networks, deep belief networks, and convolutional neural networks which are applied in areas such as comput-er vision, NLP, speech recognition, machine translation, where they help to produce better results compared to human experts. DL is extensively used in self-driving cars to process sensory data and make informed decisions. They are used for detection of roads, footpaths, signs, traffic lights, pe-destrians, cars, obstacles, environ-ment and human actions among others. DL systems proved to be powerful tools but there are some properties that may affect their prac-ticality especially when it comes to autonomous cars.The two major concerns I can give are unpredictability and their vulnerability to being fooled.Despite their accuracy (they out-perform humans in many cases), they still are unable to generalize to situations making them less trust-worthy to be fully autonomous. Since we do not yet fully understand how they work it can be challenging to diagnose and correct them imme-diately for their poor performance in novel/unseen conditions. Thus, for now DL driven autonomous car re-search is limited to the experimental phase. I believe more advancements are needed before DL or other al-gorithms can move from the labs to the realworld.Having said that, deep learning algorithms revolutionized many re-al-world use-cases enabling them with the usage of AI. Self-driving cars, heathcare AI and movie recom-mendations are better today across various horizons. AI turns science fiction into reality with Deep Learn-ing helping it to get from present to future state.
< Page 8 | Page 10 >