Sep 20, 2019
Recently I found myself with three new open tabs on my browser and realized that there are things bigger than our imagination happening right now. The tabs were namely Google AI, Microsoft AI and AWS Machine Learning.
New developments in the way we capture and analyse data were already happening, but suggesting a course of action post-analysis of that data was a task mainly performed by humans, which is now gradually shifting its course towards tasks being performed by Machines.
We all know that using Machine Learning algorithms one can easily process a lot of data points simultaneously and fetch results faster than any human can. A lot of these data points have helped organisations decipher patterns and probabilities and are being used in areas like Healthcare, Finance, eCommerce, Computer Science, Telecommunications, Robotics, Marketing, Human Resources, Media and Entertainment, Music, Transportation, Power Electronics and even the Military. Of course, it's being used by governments too.
We will discuss the application of AI and ML in all of these areas in a few minutes. But before that, I would like for you to know how we're already attached to its use in our day to day lives and how the systems are constantly learning more and more about us, individually and collectively.
Let's get a few definitions and facts straight before we delve into how it's being put to use. Just the basics.
Artificial Intelligence (AI) - In computer science, artificial intelligence (AI), sometimes called machine intelligence, is intelligence demonstrated by machines, in contrast to the natural intelligence displayed by humans. Colloquially, the term "artificial intelligence" is often used to describe machines (or computers) that mimic "cognitive" functions that humans associate with the human mind, such as "learning" and "problem-solving".
(According to Wikipedia)
Machine Learning (ML) - Machine learning (ML) is the scientific study of algorithms and statistical models that computer systems use to perform a specific task without using explicit instructions, relying on patterns and inference instead. It is seen as a subset of artificial intelligence.
(According to Wikipedia)
A few Stats about AI and ML. I mean, what's a post about AI and ML without stats? It's like a plane without navigation controls! JK.
According to Forbes and a post from CMO by Adobe here are a few stats you must know today:
I hope the speculations above must've shed some light on the importance of AI and ML for organisations and individuals. It's needless to say that by now if we're using a smartphone we all have experienced some level of AI and ML.
Use of AI and ML is soaring high especially for enterprises, it's time for you to incorporate an AI strategy in your Marketing and Business activities otherwise things might get a bit difficult and competition may be tougher.
Alright, back to where we started from, the story is such that those three new tabs on my Google chrome were continuously raising questions in my head and surprising me by each project that I read about. Here are a few insights from my readings.
The vision - Bringing the benefits of AI to everyone.
Google says that AI is making it easier for people to do things every day, whether it's searching for photos of loved ones, breaking down language barriers in Google Translate, typing emails on the go, or getting things done with the Google Assistant. AI also provides new ways of looking at existing problems, from rethinking healthcare to advancing scientific discovery.
A few of their Projects
Using Deep Learning to Inform Differential Diagnoses of Skin Diseases
Google with it's Open Source Machine Learning platform in this project developed a deep learning system (DLS) to address the most common skin conditions seen in primary care. Their results showed that a DLS can achieve an accuracy across 26 skin conditions that is on par with U.S. board-certified dermatologists, when presented with identical information about a patient case (images and metadata). This study highlights the potential of the DLS to augment the ability of general practitioners who did not have additional speciality training to accurately diagnose skin conditions.
To render this prediction, the DLS processes inputs, including one or more clinical images of the skin abnormality and up to 45 types of metadata (self-reported components of the medical history such as age, sex, symptoms, etc.).
To evaluate the DLS's accuracy, we compared it to a rigorous reference standard based on the diagnoses from three U.S. board-certified dermatologists. In total, dermatologists provided differential diagnoses for 3,756 cases ("Validation set A'), and these diagnoses were aggregated via a voting process to derive the ground truth labels. The DLS's ranked list of skin conditions was compared with this dermatologist-derived differential diagnosis, achieving 71% and 93% top-1 and top-3 accuracies, respectively.
Imagine how easy would it be for dermatologists to accurately predict the cause for a problem and provide remedial actions using deep learning, new students can learn better and faster.
Learning Cross-Modal Temporal Representations from Unlabeled Videos
While people can easily recognize what activities are taking place in videos and anticipate what events may happen next, it is much more difficult for machines. Yet, increasingly, it is important for machines to understand the contents and dynamics of videos for applications, such as temporal localization, action detection and navigation for self-driving cars.
In "VideoBERT: A Joint Model for Video and Language Representation Learning' (VideoBERT) and "Contrastive Bidirectional Transformer for Temporal Representation Learning' (CBT), Google proposes to learn temporal representations from unlabeled videos. The goal is to discover high-level semantic features that correspond to actions and events that unfold over longer time scales.
To accomplish this, they exploit the key insight that human language has evolved words to describe high-level objects and events. In videos, speech tends to be temporally aligned with the visual signals, and can be extracted by using off-the-shelf automatic speech recognition (ASR) systems, and thus provides a natural source of self-supervision. Their model is an example of cross-modal learning, as it jointly utilizes the signals from visual and audio (speech) modalities during training.
Their BERT model has shown state-of-the-art performance on various natural language processing tasks, by applying the Transformer architecture to encode long sequences, and pretraining on a corpus containing a large amount of text. BERT uses the cloze test as its proxy task, in which the BERT model is forced to predict missing words from context bidirectionally, instead of just predicting the next word in a sequence.
Google's results demonstrate the power of the BERT model for learning visual-linguistic and visual representations from unlabeled videos. They say that they find their models are not only useful for zero-shot action classification and recipe generation, but the learned temporal representations also transfer well to various downstream tasks, such as action anticipation.
An End-to-End AutoML Solution for Tabular Data at KaggleDays
Google's AutoML efforts aim to make ML more scalable and accelerate both research and industry applications. Their initial efforts of neural architecture search have enabled breakthroughs in computer vision with NasNet, and evolutionary methods such as AmoebaNet and hardware-aware mobile vision architecture MNasNet further show the benefit of these learning-to-learn methods. Recently, they applied a learning-based approach to tabular data, creating a scalable end-to-end AutoML solution that meets three key criteria:
To benchmark, their solution, they entered their algorithm in the KaggleDays SF Hackathon, an 8.5 hour competition of 74 teams with up to 3 members per team, as part of the KaggleDays event. The first time that AutoML has competed against Kaggle participants, the competition involved predicting manufacturing defects given information about the material properties and testing results for batches of automotive parts.
Despite competing against participants that were at the Kaggle progression system Master level, including many who were at the GrandMaster level, their team ("Google AutoML') led for most of the day and ended up finishing second place by a narrow margin, as seen in the final leaderboard.
After the competition, Kaggle published a public kernel to investigate winning solutions and found that augmenting the top hand-designed models with AutoML models, such as ours, could be a useful way for ML experts to create even better-performing systems. As can be seen in the plot below, AutoML has the potential to enhance the efforts of human developers and address a broad range of ML problems.
The vision - "Innovation is what creates tomorrow'.
"The Future Computed: AI and Manufacturing' is the next book in Microsoft's "The Future Computed' series looking at the impact of AI on society'.
The book features stories from industry leaders and policymakers from around the world, sharing insights into how customers can progress their AI journey. It also sets out suggestions about how countries can build competitive manufacturing sectors while ethically delivering AI and addressing short-term disruption.
It identifies six key learnings:
A few of their projects
Carlsberg is bringing AI to one of the world's oldest industries with advanced sensors and analytics that help map complex flavor profiles.
The Beer Fingerprinting Project will help researchers at Carlsberg, the fourth-largest brewing company in the world with 140 beverage brands in 150 countries, use advanced sensors and analytics to more quickly map out and predict flavors.
Known as "Dr. Beer,' Jochen Forster is the director and professor of yeast and fermentation for Carlsberg Research Laboratory, whose principal task, as laid out by founder J.C. Jacobsen, is to develop as complete a scientific basis as possible for malting, brewing and fermenting operations.
The lab began working with Aarhus University, Denmark's leading research institution, to develop sensors; with the Technical University of Denmark, north of Copenhagen, to figure out how to implement them in different fermentation scenarios; and with Microsoft to analyze the signals from the sensors using AI solutions, including machine learning algorithms, to measure the flavors and aromas created by yeast and other ingredients.
The three-year project began in 2018, so it's too early to give detailed results. But the sensors can already differentiate between various pilsners and lagers, and researchers are now fine-tuning the system and developing software that will make it easier for technicians who might not be familiar with AI to use it to amplify their work, Forster says.
The goal is to map a flavor fingerprint for each sample and reduce the time it takes to research taste combinations and processes by up to a third, to help the company get more distinct beers to market faster.
NHS Glasgow & Clyde uses Microsoft Artificial Intelligence to transform care for patients
NHS Glasgow and Clyde are trialling a pioneering use of AI to spot trends in patients suffering from chronic conditions such as Chronic Obstructive Pulmonary Disease (COPD). With each unplanned hospital trip costing in the region of 6,000 pound, the technology can play a huge part in reducing the NHS budgets required to manage people with long term conditions.
The headline goals for the program, which trials with 400 patients, have significant aims:
Although this trial focuses on COPD, it could prove to be equally valuable to other long-term conditions such as diabetes and cancer. Dr Carlin summarises "You're able to see individual benefits, you're able to see an ability to deliver treatment to patients in a realistic fashion, you're seeing improvements in quality of life, you're seeing positive changes for the NHS and also you're seeing that you're moving towards the future vision."
How the City of Los Angeles is enriching citizen and employee experiences with Chip the chatbot
The City of Los Angeles is focused on enriching the lives and experiences of constituents and improving their relationship with the city government. The city's Information Technology Agency developed Chip the chatbot, a digital personal assistant, to help residents more easily navigate government resources and procedures and free up employees to focus on higher-value activities.
Hosted on Microsoft Azure, the chatbot is currently configured to serve LA's business community and has already demonstrated a wealth of capabilities that are transforming the experiences of residents and employees alike. With Azure's reliability and flexibility, the City of LA is poised to build on Chip's success and make use of chatbot technology across the entire organization.
In line with its commitment to citizens and government employees, LA's Information Technology Agency has developed the City Hall Internet Personality (Chip) chatbot to provide city residents with cutting edge digital service which aims to achieve the following goals.
Moving forward, the City of Los Angeles identified four key priorities to continue developing their digital transformation:
The city's focus on using Chip and cloud technology to better serve residents has made the chatbot a major accomplishment and will continue to inspire its growth.
Amazon says more organizations choose AWS for machine learning than any other cloud. Their AWS offers the broadest and deepest set of tools for your business to create impactful machine learning solutions faster.
A few of their projects
Formula One uses Amazon SageMaker to analyze race data in real-time, then shares insights with television viewers.
Formula 1 is a data-driven sport: During each race, 120 sensors on each car generate 3 GB of data, and 1,500 data points are generated each second.
Using Amazon SageMaker, Formula 1's data scientists are training deep-learning models with 65 years of historical race data to extract critical race performance statistics, make race predictions, and give fans insight into the split-second decisions and strategies adopted by teams and drivers.
By streaming real-time race data to AWS using Amazon Kinesis, Formula 1 can capture and process key performance data for each car during every twist and turn of the Formula 1 circuits.
Then, by deploying advanced machine learning via Amazon SageMaker, Formula 1 can pinpoint how a driver is performing and whether or not drivers have pushed themselves over the limit.
By sharing these insights through television broadcasts and digital platforms, Formula 1 allows fans access to the inner workings of their favorite teams and drivers. Formula 1 has also selected AWS Elemental Media Services to power its video asset workflows.
Marinus Analytics uses Amazon Rekognition to find human trafficking victims who may have changed their appearance, then help to prosecute the traffickers.
Marinus Analytics uses artificial intelligence, such as Amazon Rekognition, to provide agencies with tools that assist them in identifying and finding victims, such as Traffic Jam, which assists those working on human trafficking investigations. Rekognition powers Traffic Jam through a facial recognition feature called FaceSearch.
With this tool, investigators can now save invaluable time by using image analysis to automatically search through millions of records in seconds. This is a significant improvement compared with methods investigators would have to use without Traffic Jam.
Another way Marinus Analytics uses Amazon Rekognition to aid in the fight against human trafficking is by using the DetectText API. This function can detect text in images, extract them, and convert them into machine-readable text.
Since millions of human trafficking ads on the internet have overlaying text on the image, this feature can extract all of that text, which can then be collated and organized so investigators can more easily search through it.
Although the fight against human trafficking can seem daunting, Marinus Analytics has already seen success stories.
GE Healthcare uses Deep Learning on AWS to improve the accuracy of x-ray imaging procedures while also lowering readmission rates.
The company launched the GE Health Cloud in the United States to provide radiologists and other healthcare professionals with a single portal to access enterprise imaging applications (e.g., PACS) to view, process, and easily share images and patient cases.
Further GE Health Cloud offerings, both services and applications, are in the pipeline for release this year, ranging from device protocol management to care pathway analytics.
The GE Health Cloud runs on Amazon Elastic Compute Cloud (Amazon EC2) instances. Close to one petabyte of medical imaging data is stored on Amazon Simple Storage Service (Amazon S3). The company relies on Amazon Aurora as its database service, and it uses the AWS Service Catalog to create and manage IT services.
GE Healthcare also takes advantage of Amazon Cognito for federated single sign-on to the Health Cloud for customers.
Wow, take a deep breath and relax a bit, if it was too much information all at once.
If you think you have a fair idea about what's happening in the area of AI and ML by now then you're probably wrong because these were just some of the recent works of the biggest players in the game.
There's a lot more happening in the background, from small start-ups to large scale corporations, everyone at some point is bound to use it or are already using it.
All of this comes at a cost, the cost is data and most of it is your personal data. Regulating the use of personal data is highly necessary and its use should solely lie upon personal discretion. Thus, the use of data in AI and ML should be based on values and principles which is extremely significant so that it can be put to good use.
All in all, there are great developments happening and most of them are for the good.