A Simple Approach – Efficient data, SQL query, report if / often ends up working
This is the best way to do Machine Learning (ML) and sometimes machine learning shouldn’t be done at all. In fact, according to Amazon Applied Scientist Eugene Yan, “The first rule of machine learning is to start without machine learning.”
What?
Yes, it is great to get rid of such ML models which are difficult to do after many months of hard work. And it’s not a very effective strategy. Not when there are simple, accessible ways.
It may be an exaggeration to say that “data scientists often do arithmetic,” as data scientist Noah Lorang did years ago. But that’s not far off, and we want you to know no matter how complicated the process of working with the data is, he and Yan are certainly right that it’s often best to start small.
too complicated
Data scientists are paid more. So wrapping things like complex jargon and substantial analysis in in-depth models is probably worth trying to justify that pays off. No. Lorang’s view of data science is as true today as it was a few years ago: “Machine learning has a very small subset of business problems that are best solved; most of them require good data understanding.” What is required and what does it mean. SQL queries for data retrieval, .. Lorang suggests simple methods such as basic arithmetic of that data (mathematical differences, percentages, etc.), obtaining results, and [writing] paragraphs of interpretation or recommendations. ”
I do not recommend that it be easy. I mean machine learning is not where you try to take insights from the data. Not necessarily a huge amount of data. In fact, as Able CEO Caitlin Gleeson puts it, “It’s important to start with small data [because] eye-sight anomalies led me to some of my best discoveries.” Sometimes it may be enough to design the distribution to check for obvious patterns.
Yes, that’s right: the data is “small enough” that one can spot patterns and discover insights.
[Download the New CIO Road Map Report on 5G in Business! ]
It’s no surprise to advise iRobot data scientist Brandon Roarer: “When you have a problem, build two solutions — a deeply Bayesian transformer running on Multicloud Kubernetes and a SQL query built on a layer of exaggerated hypotheticals.”
not far from here. An argument against starting in ML. To find out why in depth, it’s worth checking out Yan’s article on the subject.
people recognize data
First, Yan said, it’s important to recognize how difficult it is to make sense of data with key elements: “You need data. You need a strong pipeline to support your data flow. And above all else.” You need high quality labels.”
In other words, starting with the ML model in trouble may not be particularly useful because the inputs are confusing. At that point, you know your data. Try to solve the problem manually or with a heuristic (practical way or shortcut). Yan cites this from Hamel Hossain, machine learning engineer at GitHub: “This is a very important first step that will force you to know the problem and the data.”
Assuming you are talking about table data, start with simple interactions to start with one type of data to perform statistics and visualize the data using scattered locations, ie he said. For example, Yan argued that instead of building a complex machine learning model for recommendations, you could “recommend products that outperformed from previous seasons,” and then look for patterns of results. It allows the ML practitioner to become familiar with his data.
When is machine learning necessary or appropriate?
According to Yan, maintaining your non-ML heuristic system is what makes machine learning worthwhile. In other words, “So you are a non-junk ML.
No hard science info on when this happens
Yes, getting started with complex ML models is tempting, but one of the most important skills for a data scientist is common sense, knowing when to reverse analysis or rely on certain reports.
Artificial Intelligence is Good at Less Exciting Military Roles Too
I have received a lot of feedback on my last few paragraphs on new applications for Drones, Simulation and Artificial Intelligence. My recent column about creating an AI that can do wonders like controlling meteorological satellites performing the massive intelligence tasks found in bee colonies has generated a lot of buzz, in particular.
Most of the feedback I received was from federal companies and IT companies working on future AI projects. I will shed some light on some of these in the future. But a company called Hypergiant also has a reference to how they work by using AI to assist with unforeseen missions, not just some advanced warfare program in the Army’s Robotic Combat Vehicle program. Maintenance.
Founded in 2018, Hypergiant is an AI company dedicated to developing world-changing technology to solve the world’s biggest problems in the aerospace, defense and critical infrastructure sectors.
Sustainability is very important to the military, and it is a good example of how AI can be used to improve performance today. It may sound very common, but using AI for work tasks can have a big impact in the long run. I spoke with Quentin Donnelly, general manager of space and security at Hypergiant Industries, about how AI and other technologies like machine learning can be used to enhance performance and improve performance in military systems.
Donelan: Our goal is to improve vehicle maintenance operations. Our men and women waste so much time on the most difficult, most manual and most inefficient maintenance tasks. The success for us is that military navies are used more efficiently, and our men and women in uniform are engaged in more intellectual missions, which leads to a more efficient and effective force. on security.
NEXTKOV: While not liking the AI-like title within the autonomous military vehicle, is there an AI that controls the maintenance of those navies still a long way from ensuring their longevity and effectiveness?
Tonelen: That’s right. There is an urgent need for autonomous vehicles to use AI to respond to environmental stimuli and perform real-time actions such as avoiding obstacles or directing inclined objects. But over time, AI and machine learning can be used to find patterns in the data. Those patterns can be related to the performance or physical condition of the vehicle, and if so, they can be used to report a maintenance schedule.
NextCove: And why is it so important to keep predictability when it comes to the safety of autonomous vehicles?
Donelin: So the obvious answer is that by predicting very accurately whether a critical component will fail, you can directly reduce the number of vehicle failures. If we are talking about drones, it means fewer accidents, which will lead to fewer injuries. If we are talking about passenger vehicles, the forecasting system can detect tire malfunctions and reduce the chances of a serious accident.
Autonomous vehicles are not of particular importance here, but there is a financial aspect that I think is very difficult to attack with autonomous vehicles. These vehicles can operate in areas that are difficult to navigate or are sparsely populated. Equipment failure may require costly human intervention, which cannot be resolved in the long run, so predictive maintenance can actually reduce the number of human repair personnel. On site to repair the broken robot.
Donelan: It’s really different in every case. However, in general, a maintenance system such as a vehicle sensor has two layers that collect information such as pressure, temperature and voltage. That data can then be processed by the ML algorithm running on the vehicle, sending alerts if necessary to the local operator for immediate response. That data is sent to a center where various ML algorithms look for long-term trends.
At the highest level, data is collected from all vehicles in a consortium and compiled into a cloud-based data store, and operators can exaggerate what you create on an aggregated basis. This involves pre-ordering new or finished parts in response to computer intelligence.
Nextgov: In terms of auto parts, will the maintenance of AI have an impact on the military supply chain?
Donelin: Yes, the truth is that the stability of the forecast has the biggest impact on the business. Furthermore, if you set up a model for forecasting supply chain requirements based on vehicle condition, we can simulate a number of scenarios for supply chain demand based on naval advance conditions. For example, we may wish to simulate the phenomenon of vehicles experiencing very hot environments for a long period of time. Will this affect my projected model, if so, what does it mean from a supply chain perspective? These types of approaches are incredibly powerful.
Donelin: This is one of the most challenging aspects of machine learning because the models that generate predictive intelligence are only as good at what they are trained for. Equipping models with precisely labeled data and replacing old models with new ones is an ongoing battle.
Updating the ML instructions on a vehicle can be confusing because those vehicles may not always be connected to the Internet or the network to which software developers send updates. Everything we do is focused on creating a digital infrastructure that allows ML engineers to retrain and refactor models, even at the edge of runners or used computers.
