Our developed world contains 'infrastructure' - buildings, freeways..., overlaid on to the 'natural' environment.

A variety of data exists, regarding these - collected by governments, academia, industry. So, we could/SHOULD mine them.

What if we could predict traffic congestions, using past data? An app that can do this, Waze-style, would help.

Rather than just use real-time data (Waze-style), we could ALSO use historic (past) data, and, spatial dependencies (the structure of roads and freeways).

Specifically, data from previous events will help modify current routes (assuming that events behave identically).

One of USC's projects does this, using data provided by LA.

ClearPath is the name of the app that resulted from the research. ClearPath also aims to optimize routes for fleets [FedEx, UPS...], and even public transportation [buses].

This is another writeup of the USC research.

What if we could store past traffic flows as images, do a match with the current flow, and use past continuation patterns to route current traffic?

This is a toy example, fully worked out - the idea is to 'learn' commute times, and use it to predict commute times for future drives.

There are 2 different/complementary ways to collect traffic data.

Sensors embedded in our freeways help create realtime traffic maps.

When we use Waze, our phones act as sensors - they relay OUR GPS location to Waze, which uses COLLECTIVE data to possibly REROUTE traffic! Here is what Waze sees right now..

So the difference in the above two is this: Google map shows data collected from sensors, while Waze shows data from users' phones - this means that the Waze map can have 'gaps', in places where there aren't Waze users driving.

Given that violations do occur, how best to allocate enforcement resources? Past data can help build 'predictive models'.

Here is a DSSG project, for NY data; this is another description.

Here is a mapping (viz) project - pollution levels in San Diego. It uses R code [look in the GitHub repo].

'Sustainable growth' is the goal/dream, for cities everywhere - spatial data analysis plays a crucial role here, as shown in this ESRI clip.

Data plays a HUGE role in managing realtime emergency responses (eg during natural disasters such as wildfires).

Learning buildings' energy usage data can lead to intelligent forecasting of future energy use...

It is also useful to be able to predict electricity generation and usage [includes a long primer on ML].

Here are six applications, all energy-focused.

We generate MOUNTAINS of trash, collectively - so it makes sense to use DM/ML to deal with it intelligently.

This paper analyzes the use of 3 different ML methods, to sort trash [at a recycling facility].

It is useful to be able to predict waste (trash) generation, eg. to properly utilize trash collectors and vehicles. Here is one approach that uses a neural network. This book presents MANY DM applications - pages 166-180 present a hybrid approach involving clustering and classification, for municipal waste prediction.

Here is a way to use Torch [a popular ML alternative to TF, esp. in PyTorch form] to classify --- garbage :) DO take a look at the training+testing dataset of images, in 6 categories, and simple text files that contains the labeled data ('image trash_type' pairs) - click on 'constants.py', and 'one-indexed-files-notrash_train.txt', to see how the labeling is done, then download the .zip file with the pics of training+test images, look through them to see the variety of 'trash' images used for training.

Water is a VERY precious resource, one that enables life on earth!

This is an approach to using ML, for intelligent distribution of water.

'Hydroinformatics' - data science meets water - this is a good summary of the various ML/DM algorithms in use.

Here are many types of 'smart water' ['digital water'] applications, all focused on efficient, responsible management and consumption of water.

"Everyone complains about the weather, but no one does anything about it" - at least we can predict it well :)

NOAA, the country's "weather people", use ML for predicting impactful weather events [in addition to routine weather prediction they have been doing for decades].

A variety of private companies are also into ML-based weather forecasting.

Climate - even 'bigger' than weather - is an important factor in determining our PLANET'S FATE.

This is what NASA has to say [news flash - it's not "fake news"!].

Here is an NIH document, on the 'data' aspects of climate change.

Skeptics' uninformed opinions aside, climate change is REAL (there is a reason why both hurricanes AND wildfires have been getting bigger lately), and ML can be used to address it.

Electronics/tech have always been a huge part of US' successes in wars; now we can add AI to the mix to create 'intelligent warfare'. This article lists some areas where AI/ML can help.

The NSA, among other things, needs to sift through mountains of data, to glean actionable knowledge. Here is a typical data science-related job description.

Here is an 'automated security guard' [said to be deployed in the DMZ between the Koreas].

Terrorism in the future might not involve 'direct' operations [shooting, suicide-bombing etc] - here is a terrifying scenario of what could be.

We have brought up self-driving cars, a lot, during this course.

It would be highly beneficial to train, and test, autonomous cars in a virtual environment. Benefits:

• no RL accidents/fatalities!
• a HUMAN can observe the driving
• extreme hazards can be programmed in
• pause-and-rewind!
• 100% access to all the data, eg. to reconstruct an accident

Quake prediction would be nothing short of amazing - here is one approach, using random forests (a group of decision trees). Even predicting aftershock locations would be quite helpful.

The Yield project by Microsoft is about hyper-local weather precictions for farmers.

Beck's Hybrids uses ML to predict the best seed to cultivate. Nutonian is the company that supplies the DS solution.

ML can be used to identify bugs in crops.

Here, 'good old' TensorFlow is used to train an NN on diagnosing cassava plant diseases, and deployed via a smartphone.

Here is a way to automate harvesting [and reduce pesticide use, track cow behavior].

Here is more - harvesting, spraying, monitoring, analyzing...

Manufacturing is, loosely speaking, assembling/transforming raw materials into finished products, using a series of steps. ML is useful in practically every aspect of this:

• supply chain management
• inventory management
• asset management: Overall Equipment Effectiveness [OEE] for tracking availability, performance and quality in order to reduce downtime and inspection/testing costs, and to increase yield [eg. watch the 'Hirotec' video here]
• performance of manufactured product

This note discusses ways in which judicious use of data can aid manufacturing.

IIoT - Industrial Internet of Things - is the name given to the idea of using sensors attached to equipment, raw material pallets, even manufactured products - to enable data collection and analysis.

This page contains case studies from manufacturing giants Siemens and GE, and robotics giants Fanuc and KUKA. Here is another page with related content.

Here is a Forbes page that lists 10 ways in which ML is helping manufacturing.

A 'digital twin' is a virtual replica, of a RL product. Why would a manufacturer build one? To streamline:

• design
• manufacturing
• product use
• maintenance

Data is what connects the real and virtual counterparts. Here is a paper on digital twins, and this brief discusses how a company can get started on this.