Machine Learning (ML) and Artificial Intelligence (AI): Prominent ML & AI applications including those on Mobile devices: Part - Twelve

Dr. RGS Asthana

Life Member IEEE

Figure 1: funding and Mobile app [17]

Summary

Mobile Technology, Big Data, 3D Printing, Self-driving Cars (SDCs), Robotics, Drones, Speech Recognition & Machine Translation, video processing and recognition are very important technologies w.r.t. ML and AI. We briefly describe prominent technologies, such as, reinforced learning, generative models, networks with memory, learning from less data and building smaller models, latest ML and AI Hardware and also simulation environments.

ML technology based applications like exploitation of Big Data and Data Mining, finance, 3D printing, robotics, drones and e-commerce. Self-driving cars and also their intelligent use in smart phones is on the increase. We also briefly talk about quantum computers in the end of this article as it appears to be future technology. Quantum computers are many times faster than traditional computers in certain type of operations.

Prerequisite

Read articles [1] to [17]

Keywords

Machine Learning (ML) Tools, Artificial Intelligence (AI), Neural Networks, Internet of Things (IoT) and Deep Mind.

Prelude

ML, AI, Mobile Technology, Big Data, 3D-printing, Self-driving cars, Robotics, Drones, Speech Recognition & Machine Translation, Video Processing and Recognition of Objects are playing significant role. What really marks healthcare different from other disciplines? AI is going to be smarter than us and will be much faster than what we think. AI will write software itself. Healthcare may often have very little labeled data (e.g., clinical NLP). This may prompt the use of semi-supervised learning algorithms i.e. keeping human in the loop (HITL). Sometimes, we have only small numbers of samples (e.g., for a rare disease) and we need to learn as much as possible from other data (e.g. EHR data of healthy patients). We may have lots of missing data that too at varying time intervals and may only get censored labels. Other more important problem which we need to solve is that ML base algorithms do not give reason for arriving at a particular decision. Therefore, it is pertinent to model the problem keeping these aspects in view and may be reason for HITL in the solution.

Power of AI is analyzing images faster than human surveillance camera with AI software. Further, drones can recognize faces and people from their ear pattern, movement from top. Same is true for robotics as AI software is behind them giving them all the power.

ML based solutions are good at prediction and diagnosis too is a prediction in a way. We, therefore, describe ML based diagnosis and treatment systems. The only thing necessary for systems to give better prediction is training on substantial data. The areas where ML/AI based systems have impact in healthcare are: on-line consultations, Health assistance and medication management, Personal genetics, development of drugs of the future, discovering new diseases, persistent care, discovering new clinical pathways and last but not the least Robotics and Healthcare.

ML mobile app development is the next big thing. Google introduced software that uses neural nets and provides language translation. Technology is being optimized for smartphones and may also work without any internet connection. Lenovo new smartphone can run number of apps without Internet connection. Apple SIRI SDK [25] is now accessible to all developers and they can easily incorporate SIRI features into their apps. Mobile smart phones are with hundreds of millions of people all over the world. It is therefore, logical to use ML based apps on these phones for maximum reach.

A mechanism is established to collect data while the system is running.

Areas of Prominence and Research [34]

Reinforcement learning (RL)

RL is much like animal or human learning i.e. giving rewards for every right move and penalty for every wrong move. The idea is to optimize the right moves till we reach the goal. Applications include multiple agents either learning in their own instance of an environment with a shared model or by interacting or learning from one another in the same environment, how to navigate 3D environments like mazes or city streets for say for autonomous driving, inverse reinforcement learning to recapitulate observed behaviors by learning the goal of a task.

Generative models

This is a branch of unsupervised learning. By sampling from this high-dimensional distribution, generative models output new examples that are similar to the training data. For example, a generative model trained on real images of faces can output new synthetic images of similar faces. Generative Adversarial Networks (GANs) offer a path towards unsupervised learning. With GANs, there are two neural networks: a generator, which takes random noise as input and is tasked with synthesizing content (e.g. an image), and a discriminator which has learned what real images look like and is tasked with identifying whether images created by the generator are real or fake. Adversarial training can be thought of as a game where the generator must iteratively learn how to create images from noise such that the discriminator can no longer distinguish generated images from real ones. Applications of this technology include Simulation of possibilities of a time-series; recovering 3D structure from a 2D image [36]; generalizing from small labeled datasets [37]; creating natural language in conversational interfaces; cryptography [38]; and Synthesizing music and voice [39].

Networks with memory

It is considered necessary that the system must learn continually new tasks and remember how to perform all of them into the future just like human being. However it is also important that the networks have long term and short term memory to remember weights of each node of ANN for use in the future.

These include long-short term memory [40] networks - a variant recurrent neural network that are capable of processing and predicting time series; DeepMind’s differentiable neural computer [41] that combines neural networks and memory. Applications Include Learning agents that can generalize to new environments; robotic arm control tasks; autonomous vehicles; time series prediction (e.g. financial markets, video, IoT [17]); natural language understanding and next word prediction.

Learning from less data and building smaller models

Deep learning models require huge amounts data for training. This data requirement grows when a single ANN is used e.g. taking raw audio recordings of speech as the input and outputting text transcriptions of the speech.

An another approach may be to develop learning of a new task by using knowledge of a machine learning model acquired from a previous task using processes collectively referred to as transfer learning [42] e.g. if we have model to recognize cars it could be easy to modify it for trucks. Applications include text classification, spam filtering and machine translation.

ML and AI Hardware

Recent progress in AI can be attributed to the use of graphics processing units (GPUs) for training large ANN models. Training on GPUs is much faster than with CPUs. The first ANN implemented on a GPU. NVIDIA continues to lead the charge into 2017, ahead of Intel, Qualcomm, AMD and more recently even Google. The accuracy with which GPU operates may not be needed in ML tasks so Apple is developing a chip called the Apple Neural Engine (ANE) for its mobile devices [32].

Today AI based digital assistants include in machine speech-recognition capability, for example, Siri from Apple), Google Assistant from Google, Cortana from Microsoft, Bixby from Samsung, and Alexa from Amazon.

Augmented reality and digital assistants are not the only applications of AI that are significant on mobile devices but there are many more to come. For example, health applications will be able to tell when body readings from sensors on the phone or associated wearable devices are abnormal and need acting on. These devices may even work when mobile is not on Internet.

It is not a must to have chip for neural processing. Chip-maker Qualcomm for example has provided a software-based approach, called the Snapdragon Neural Processing Engine [33], to allow developers using their chips to incorporate AI into their software.

However, an AI chip is the latest inclusion in the latest smart phones [47]. AI hardware speeds up the process ML and AI apps, e.g. Google's AI hardware, called Tensor Processing Unit (TPU), is 15 to 30 times faster than the fastest computer processors (CPUs) and graphic processors (GPUs) that power computers today. Applications include running AI systems at the edge (IoT devices [17]); always-listening IoT devices; cloud infrastructure as a service; autonomous vehicles, drones and robotics.

Simulation environments

In fact, in real world one desires to develop a general purpose AI system. These environments present raw pixels to an AI, which then take actions in order to solve for the goals they have been set (or learned). A list of simulation environments is given in [43].

ML application areas

Machine learning is a very multidisciplinary field and can find its implementation at the intersection of technologies, science, and business.

Big Data and Data Mining

The ML success depends on data labeling and analysis. ML Algorithms, in fact, are a refined data analysis method as they offer a deeper insight into collected data and help find hidden patterns. The foremost goal of ML when used on big data is to identify useful patterns in it and ML algorithms predict future patterns too. ML applications are based mainly on data to develop more predictive models.

ML can be applied also to smaller datasets; however, the outcome may not be very accurate as learning chance is low. ML over the years has evolved from simple analytical algorithms to automatic application of customized algorithms and mathematical calculations to big data. The speed and iteration are unique qualities of ML. In some applications sometimes it is essential to diagnose things based only on individual’s test results and some doctors do it very well. That does mean the field is healthcare an area where humans may be willing to take risks if one perceives rewards in near future.

Big data [7, 12] may comprise of the collection of structured as well as unstructured data. The field of data mining assists in analyzing big data and to determine interesting links within significant set of data. It consists of the data storage, maintenance and the actual analysis. Here, ML algorithms play a big role in finding all possible relationships in the data.

Finance

In finance ML algorithms predict future trends in market, bubbles, and crashes. The system can compute an outcome for an individual considering his portfolio & his credit rating and send to his smartphone recommendations.

3D Printing [7]

3D printing will be using many materials including live tissues for organ printing. However, biggest breakthrough may come when it becomes a standard surgical procedure in hospitals. Nano technology may get a boost due to advancement in 3D printing as it may be possible to print Nano-level pieces of hardware with precision and accuracy. Let’s imagine that a shoe company like Adidas or Nike turning into a software company and selling only shoe or clothing design and requisite software to 3D print the shoe in the comfort of customer’s home. It may, however, be taken for granted that cyber security will advance sufficiently to take care of any issues. This may change mode of manufacturing also.

The first ever Smart-phone 3D Printer is available at less than US$100. It is portable and multi-material and works accurately for professionals and has very easy operation for new users. With an iPhone 6, OLO can make 1 cm object in approximately 46 minutes, or print 1 inch in approximately 1 hour and 55 minutes.

Smart phone is almost half of its hardware which is already in your pocket with OLO hardware which is only for USD 99, thus, empowering anyone to print in 3D (See figure 2).

Figure 2: OLO - under USD 100 3D printer [19]

4D printing [49] is also evolving at MIT in collaboration with ‘Stratasys’ where a single print with multi-material features can change from any 1D strand or 2D surface into 3D shape or morph from one 3D shape to another.

Robotics

More than a quarter-million Americans turn 65 every month. In US very soon, millions of people will need help as their health predictably becomes feeble with time. It costs around $90,000 a year if you hire a Skilled Nursing Facility and around $25,000 per year for hiring an Assisted Living Facility [6].

In view of this the case of elder-care robots will play an important role, i.e., not all dependence upon machines is necessarily bad. Robots could most notably help ill or disabled people, in future; these people could leave the house with help of the robots and live fuller lives.

Cognitive services are offered by all major key players in the field and they invariably use ML in the cloud [6]. Cognitive services are mainly accessible in computer Vision to extract information from images to categorize them and process visual data; and speech: speech to text as well as speaker recognition.

Today we are in the era when robots assist people on work and household, take care and entertain them. It is possible to manage Robots through smartphone.

Cloud Robotics [46] is possible today as interconnectivity reaches the land of AI also. These systems allow robots with multi-task capability to work on specific problems individually and exchange solutions among them. The robots share the data on cloud, enabling it to be examined by any other robot or intelligence system also connected to the same network. Thus, using collaborative effort thus performance of the entire system is enhanced.

Figure 3: - C3po – a Humanoid character as shown in ‘Star Wars’ episodes(https://www.google.co.in/url?sa=i&rct=j&q=&esrc=s&source=images&cd=&cad=rja&uact=8&ved=0ahUKEwjl5Pe9rtbXAhXJO48KHe-YDuIQjhwIBQ&url=http%3A%2F%2Fstarwars.wikia.com%2Fwiki%2FC-3PO&psig=AOvVaw3RlzL1pi0t8zySSMZDvN8P&ust=1511583762518414)

A robot essentially is combo of various technologies including speech recognition, visual computing and mechanical engineering, among other things. There exists some commonality with smart phone. These technologies can be collected including embodying a smart phone [50] to create a "universal" robot like C-3PO [51] – a Humanoid character as shown in ‘Star Wars’ episodes (see figure 3). This will also reduce number of components due to duplicate functionality.

Robots can be designed in any shape and size to access even that area where human cannot go. These robots can be used Nanorobots can be employed for removing the heart blocks (see figure 4) in more effective and accurate manner. Current diagnostic measures include painful processes like the angiogram [55]. The treatment for the block is also extremely dangerous, time consuming and painful. Angioplasty, although having the higher success rate, is old fashioned. Nanorobots can be used in this process of curing heart blocks. A standard idea is to inject an army of biodegradable nanobots into blood stream of a person with CAD and these robots will sweep the artilleries with cholesterol deposit and degrade with time.

Figure 4: Cholesterol deposit [55]

Emergency responders risk life and limb interacting with known hazards to protect the public, rescue potential victims, and enhance the resilience of communities. Firefighters, bomb technicians, and urban search and rescue specialists typically wear only conventional personal protective equipment while dealing with a variety of extreme hazards for which remotely operated robots should be well suited [56].

With reports that millions of jobs will be lost and taken over by robots say, by 2020, businesses that adopt and absorb automation process as well as new technologies will be able to free and use their possessions (be it manpower or funds) to develop state-of-the-art technologies.

E-Commerce

ML and ecommerce unbolt new opportunities for revenue and better customer experience [24]. Over time AI has made an impeccable space in the market. A study reveals that by 2020, around 80% of customer interactions will be handled by AI [52]. Big retailers’ viz. Amazon and eBay already exploited and showed it to the world that revenue and customer experience can be enhanced using ML. Some apps using ML are explained below:

1. Product categorization/Search [21]

2. Product Recommendation and Promotions [20]

3. Trend forecasting and analytics [22, 24]

4. Fraud detection and prevention [23, 24]

A few tips are given below for success of ML projects:

1. Big data helps increase accuracy as more and more data is provided to the algorithm the more accurate results and predictions are obtained.

The success of the ML project depends on selection of right ML algorithm.
Data scientist plays important role in the project as he chooses the right method and parameters to get to the best results.
Utilizing data features effectively enhances success of the ML project.
ML algorithms need careful testing.

Chatbots [53] can execute routine tasks for operations and marketing e.g. these software tools not only automate but also order procedures and provide customer service. If a customer is online, he may already be logged into some social platforms such as Facebook. This offers an opportunity to use messenger to be proactive and this opportunity could be used for confirm orders and providing online support, as the case may be.

We can talk to a smart phone, laptop or even a home appliance with the help of cloud-based AI software agents viz. Virtual assistants are Siri from Apple; Google Now from Google; Cortana from Microsoft and Alexa from Amazon. The latest advances in virtual assistants include Natural Language Processing (NLP) capability so they do recognize what people are saying in words or text. Virtual assistants do influence the way customers purchase.

Alexa has emerged in commerce as it has been successfully integrated into Amazon’s own products as well as products from other manufacturers. Microsoft’s Cortana and Amazon’s Alexa will be integrated” by the end of the year 2017 [54] so as to exploit the unique capabilities of each other and make it accessible to the user. For example, in an event of local sale one can book taxi from Uber and dinner from say Pizza hut.

We describe below some interesting and intelligent apps for smart phones:

Ford acquired Chariot and Israeli ML and computer vision startup SAIPS [30]. This technology brings Ford image and video processing algorithms, as well as deep learning algorithms focused on processing and classifying input signals. Drive.ai [31, 8] - a Silicon Valley startup founded by former lab mates out of Stanford University’s Artificial Intelligence Lab – it is creating AI software for autonomous vehicles using deep learning, which we believe is the key to the future of transportation. Google launched a car (see figure 5) recently which is programmed to handle self-driving uncertainties. It uses deep learning — a type of ML techniques to power everything in the car from the sensors and cameras, to the vehicle’s decision-making, to the way the car communicates with people and things around it. Apple is also working and may launch its own SDC soon.

Figure 5: Google's fully functional driverless car [12]

But what happens when a SDC is involved in an accident, or when a drone violates privacy rights? This is where ethicists, insurers, lawyers, policymakers, transport specialists and business planners play a role. There is need to sensitize people to the regulatory and ethical-moral issues associated with these new technologies, e.g., when an accident is about to occur whether a SDC will have priority to save its passenger or the pedestrian?

Drones, AI and smart phones

On-device AI is just beginning to happen, but it's going to need all the computing power it can get; this will save battery too [60]. In fact, if we look in a generalized manner a drone is flying robot therefore what applies to a robot also applies to a drone. In a smartphone instruments like gyros, accelerometers, IMUs, and high resolution cameras are standard devices and it makes them ideal for low-cost brains for robots as well as drones. In [61] as shown in figure 6, a sophisticated platform like vision-based real-time autonomous navigation of a flying robot or a drone or a Quadcopter is controlled by a very basic consumer device – a smart phone.

Figure 6: Drone and Smart phone [61]

It may be worth mentioning here that Quadcopter has only the Quadrotor, a motor controller and a battery. Other all features of Quadcopter are being provided entirely by the phone, which is just a stock Android smartphone with a Qualcomm Snapdragon inside.

Disaster Management: Rope bridges [8]

A drone or group of drones may not only provide data, but use AI and ML techniques for disaster management by physically augmenting the infrastructure around us, that will be the most cutting-edge usage of UAV technology. One such example could be that Drone sets up a rope bridge say across a river on a stormy weather day. It will however need special hardware, software, battery life, and also approval from FAA (in USA) to implement. Drones may also become a part of IoT [17].

Drones offered as a service [8]

The Skymatics app [57] is available on Google Playstore and clients can book UAV services and jobs instantly on their mobile device.

This Drone works on voice instructions as well as has capability to identify faces. It was built at a cost of less than USD 200 [58]. The Azure Face API [59] lets you upload pictures of your friends and it recognizes them. It also gives face attribute like age, emotion, gender, pose, smile, and facial hair along with 27 landmarks for each face in the image.

Way forward

AI is now one of the most popular topics in business as well science. Google’s $400 million acquisition of DeepMind is a prime example of mainstream AI application. Further, investments in AI, particularly, in the field of healthcare, education, and finance are increasing, but mobile is becoming one of the most promising areas for AI.

A chip smart enough to think on its own or one that can imitate the human brain, is today’s cutting edge but no lab projects is even close to achieve this feat [48]. In a commercial smartphone, the idea is simply looks far-fetched. These new processors help make software jobs such as ML jobs not only better-organized but also faster.

If growth in ML and AI is not monitored and controlled it may become main cause or threat to our existence although there is no stopping of AI evolution now. AI will only become smarter, faster and human-like in time to come. In Quantum computer a quantum bit or Q- bit has state ‘0’ or ‘1’ or ‘0’ and ‘1’ at the same time, i.e. Quantum super position. This feature is very useful and can really be exploited in cryptography. Think of a four binary combo ‘0’ and ‘1’ lock. In case of conventional computer one needs to try all combinations to open lock or to get the correct code. But a quantum computer can try all codes at the same time and get the correct code at the same time. Further, 128 - bit encryption or 256 - bit encryption can be broken in a matter of few seconds whereas a conventional computer can take many years to break the code. Quantum computers because of great speed will solve all mysteries of life regarding the environment, aging, disease, war, poverty & famine, origin of the universe and deep-space exploration. Quantum computing needs classic computers for the output.

IBM has announced two powerful new quantum computer processors [44], one client-ready and another in the works (see figure 7). November 2017 announcement includes both a 20 qubit processor ready for use by IBM Q clients and an operational 50 qubit prototype currently in development. Quantum computer is fast only in some operations so classical or conventional computer are going to stay. Problems where power of exponential is vital, quantum computers are very good but classical computers may fail to solve, such a problem and such problems can be solved only on classical computers if we go for major approximations. Quantum computer exploit principles of super conductivity and, therefore, need to operate at about 15 degree kelvin.

Figure 7: Quantum computer from IBM

Amazon and Netflix have improved their turnover in business by using predictive technology. Pandora recommends music to their clients by using AI. Nest uses AI based behavioral algorithms to predictively learn from your heating and cooling needs, thus anticipating and adjusting the temperature in your home or office and makes sure that you are comfortable.

It seems now that augmented reality (AR), virtual reality (VR) and AI are the evolving technologies and have great potential to change our everyday as well as business lives too. In 2016 [45], Pokémon Go and Snapchat demonstrated how brands can effectively capture consumer attention and monetize augmented reality experiences at scale. Add to this recent news from Digi-capital, which showed that in the first quarter of 2016 investment in augmented reality and virtual reality grew to US$1.1 billion whereas in 2015 this figure was only US$0.7 billion only [45].

The future of AR is in its potential to access content. Soon Microsoft’s HoloLens and Google Glass could develop necessary hardware and software for AR to check and show our emails, posting on Facebook and discovering the best route to our meeting place across town, with all content delivered straight to our eyes.

In fact today, most of output is on flat i.e. 2D screens but the real world is 3D, it is possible to visualize by using AR and VR tech in 3D. Use of AI tools along with AR and VR has and will make it more and more powerful and useful for humans in years to come, particularly, on smart mobile devices. Healthcare and Games are the kay players now.

References

[1] Progress and Perils of Artificial Intelligence (AI)

http://newblogrgs10.blogspot.in/2017/04/progress-and-perils-of-artificial_5.html

[2] Invited Chapter 6 - Evolutionary Algorithms and Neural Networks, Pages 111-136, R.G.S. Asthana in book, Soft Computing and Intelligent Systems (Theory and Applications), Academic Press Series in Engineering, Edited by:Naresh K. Sinha, Madan M. Gupta and Lotfi A. Zadeh ISBN: 978-0-12-646490-0

http://www.sciencedirect.com/science/book/9780126464900

[3] Future 2030 by Dr. RGS Asthana, Senior Member IEEE

https://www.linkedin.com/pulse/future-2030-dr-rgs-asthana-senior-member-ieee-r-g-s-asthana

[4] Machine Learning (ML) and Artificial Intelligence (AI) – Part 1, by Dr. RGS Asthana, Senior Member IEEE

https://plus.google.com/113904112609362672954/posts/4ti2gyQVPZ7

[5] Machine Learning (ML) and Artificial Intelligence (AI) – Part Two, by Dr. RGS Asthana, Senior Member IEEE

https://plus.google.com/113904112609362672954/posts/X5wB7baJLwt

[6] Machine Learning (ML) and Artificial Intelligence (AI): Cognitive Services and Robotics – Part Three by Dr. RGS Asthana, Senior Member IEEE

https://plus.google.com/113904112609362672954/posts/EA9hZcY6AwL

[7] Machine Learning (ML) and Artificial Intelligence (AI): Big Data and 3 D Printing – Part four by Dr. RGS Asthana, Senior Member, IEEE.

https://plus.google.com/113904112609362672954/posts/agAa6pY9ekA

[8] Machine Learning (ML) and Artificial Intelligence (AI): Drones and Self-driving Cars– Part Five by, Dr. RGS Asthana, Senior Member IEEE

http://newblogrgs10.blogspot.in/2017/05/machine-learning-ml-and-artificial_17.html

[9] Machine Learning (ML) and Artificial Intelligence (AI): Healthcare– Part Six by, Dr. RGS Asthana, Senior Member IEEE

http://newblogrgs10.blogspot.com/2017/05/machine-learning-ml-and-artificial_26.html

[10] Machine Learning (ML) and Artificial Intelligence (AI): Will AI/ML intelligence surpass humans? Part Seven by Dr. RGS Asthana, Senior Member IEEE

http://newblogrgs10.blogspot.in/2017/06/machine-learning-ml-and-artificial.html

[11] Machine Learning (ML) and Artificial Intelligence (AI): Impact of AI/ML in Healthcare: Part-Eight by Dr. RGS Asthana, Senior Member IEEE

http://newblogrgs10.blogspot.in/2017/07/machine-learning-ml-and-artificial.html

[12] Machine Learning (ML) and Artificial Intelligence (AI): Big data & Data Science (DS) and their importance: Part-Nine by Dr. RGS Asthana, Senior Member IEEE

http://newblogrgs10.blogspot.in/2017/08/machine-learning-ml-and-artificial.html

[13] Machine Learning (ML) and Artificial Intelligence (AI): Super-Intelligence - Are we afraid?: Part-ten; by Dr. RGS Asthana, Senior Member IEEE.

http://newblogrgs10.blogspot.in/2017/09/machine-learning-ml-and-artificial.html

[14] Machine Learning (ML) and Artificial Intelligence (AI): ML Algorithms: Part- Eleven

http://newblogrgs10.blogspot.in/2017/10/machine-learning-ml-and-artificial.html

[15] Deep mind website

https://deepmind.com/

[16 IBM Watson Website

https://www.ibm.com/watson/

[17] Internet of Things (IoT)

http://www.cisco.com/c/en_in/solutions/internet-of-things/overview.html

[18] How to use ML in Mobile