Analyse prédictive : prédire ou périr?
Inaugurated on June 3, 2019, the Joliot-Curie supercomputer hosted by the French Alternative Energies and Atomic Energy Commission CEA) can carry out 9.4 million billion operations per second at full capacity.
The supercomputer is used for such things as predicting weather changes and analyzing air turbulence inside aircraft engines (to reduce noise at takeoff), and its petaflop computing power (1 petaflop = 1 million billion floating point operations per second) enables it to perform ultra-complex modelling using huge volumes of data.
The supercomputer uses Intel Xeon 8168 2.7 GHz processors with 24 cores per processor, and Intel Xeon Phi 7250 1.4 GHz processors with 68 cores per processor. That makes a total of 79,488 computing cores for a computing power of 6.86 petaflops, and 56,304 cores for a computing power of 2.52 petaflops, respectively.. Imagine 75,000 desktop computers humming in unison over 600 square metres.
The investment made by France (and, by extension, Europe, the U.S. and China, the main leaders in a global market worth around $10 billion per year), demonstrates that strategic issues in high-performance computing and data analytics are crucial. Computer simulation and hyper-intensive computing for predictive purposes have become essential tools for the advancement of knowledge and basic and applied research in a growing number of sectors.
Most sectors, including aeronautics, oil and energy, epidemiology and medical research, prevention of natural disasters and insurance, and even large-scale electronic marketing and management (profiling?) of users of digital social platforms cannot do without supercomputers to direct their innovation strategies. Their objective: to build models that use past events, trends and behaviours as a basis for extrapolating and predicting events, trends and behaviours that are yet to come.
A good example of high-performance data analytics at work is the self-driving car, which must assimilate huge volumes of data to manage its performance on the road in real time. This example also brings to light certain real‑world constraints – and their associated risks.
Assisted driving or self-driving cars: society’s choice
Two options exist:
- The vehicle provides driver assistance in the form of automatic parallel parking and integrated automatic guidance systems.
- The vehicle drives itself. In this case, the vehicle has to predict all the things that might happen on the road, as a human would.
To do this, the vehicle has to:
- manage the multitude of data it captures from the environment (pavement markings, road signs, signals by persons directing traffic, illuminated arrows and cones at construction sites, road barriers, etc.) and understand what it means;
- make decisions (maintain a safe distance, determine the appropriate braking distance in snow conditions, yield, make a detour if possible, stop, slow down, recognize a false positive, etc.) based on accurate interpretation of the data that is collected; and
- anticipate the behaviours of other road users (vehicles, pedestrians, motorcycles, cyclists, users with reduced mobility) and “occasional users” (deer, moose, groundhogs) and react, if necessary.
In addition to using processors, software components and learning algorithms, self-driving cars perform such operations with the help of four types of sensors:
- cameras for seeing what is around them;
- LIDAR Laser Light Detection and Ranging) laser systems, which produce three-dimensional images of objects in the environment;
- radar, which enables the car to calculate the distance between itself and other vehicles using radio waves; and
- sonar, which detects nearby objects using ultrasound.
Although self-driving shuttles and cars have been tested in Canada and elsewhere with impressive results, these vehicles raise issues that go far beyond automated data management in the field of transportation.
In 2018, a woman walking her bicycle across a street was hit by a self-driving car being tested by Uber at 60 km/h. She died from her injuries shortly afterwards: More recently, a team from Radio-Canada raised the stark question of whether self-driving vehicles could be made to decide who should live and who should die in the event of a brake failure or inevitable collision. Can we trust these vehicles’ reactions in emergencies? Can we even agree on how they should react?
It’s not surprising that 84% of respondents to the Evolution of Mobility: Autonomous Vehicle survey said that they would prefer to drive their cars themselves, even if the car had self-driving features.Car manufacturers will have to consider the low social acceptability of self-driving cars in their promotional campaigns.
Profiting from predictive analytics
Despite the well-founded concerns surrounding self-driving cars, predictive analytics has many applications that cannot be ignored if organizations want to perform and innovate, including:
- inventory management and optimization;
- calculation of cash flow;
- talent planning and, more generally, talent supply chain management;
- risk and compliance management;
- evolution and maintenance of IT systems;
- proactive management of customer relationships (customer intelligence);
- automated testing that allows organizations to increase conversion rates or detect gaps in quality;
- detection of anomalies through a preventive maintenance approach; and
- identification of the target group for a marketing campaign and of the best communication channels for effective contact with such groups.
If properly executed, predictive analytics allows organizations to overcome the physical constraints that might prevent them from testing scenarios. This is because predictive analytics can be used to digitally simulate how something or someone functions in a controlled environment. When it comes to developing a business, or new products and services, who can say no to predictive analytics?
Imagine you sell pools and garden furniture.
- Can you establish a reasonably reliable statistical correlation between a type of product to be promoted among existing or potential clients, and weather forecasts for the following summer?
- Can you use your data to establish a relationship between a type of pool (above-ground, in-ground), a neighbourhood, the average purchasing power of that neighbourhood’s residents, and the profile of consumers who can be effectively targeted?
- Is the quality of your data (freshness, accuracy, coherence, intelligibility, relevance) high enough for you to make the right offer to the right customer at the right time?
- Would you know what to say right now?
Predictive analytics makes all of the above possible. Better yet, it confirms the old adage that a good businessperson makes sure that people buy what they need to sell