University courses for physics, artificial intelligence, neural networks, AI, computer systems design.
Professional courses in AI and artificial neural networks. How to upgrade your business, cloud computer and artificial intelligence.
School and University The prestigious MIT offers hi-tech online courses
Accessible online to all students of the world.
Commendable initiative of the prestigious MIT announcing the MITx program to attend free courses via an open source platform. Accessible via the Web to all students in the world who want to learn.
“The courses are free but the official certificate is paid”
Luxury e-learning – Attending a course at the Massachusetts Institute of Technology costs very expensive, not only for tuition but also for room and board. Not everyone has the economic opportunity to move to Boston to attend classes live. However, many dream of an education of that level. And now it’s possible and free. It’s not yet about the possibility of remotely graduating from MIT, but it’s a start. Those who wish, in fact, can attend an interactive course through the OpenCourseWare e-learning platform which includes lectures, online laboratories, self-assessment tests and discussions among students.
Official Certificate – Susan Hockfield, president of the Massachusetts Institute of Technology, explains that anyone with motivation and commitment should be able to experience their courses. And that their goal is to develop new approaches to online teaching. It starts with a basic course from next spring, free and for everyone, which also provides for the issue of an official certificate for those who have attended it successfully, and for a small fee that will go to finance the MITx project.
From MIT artificial neural networks that adapt to constantly evolving data streams
Researchers at the Massachusetts Institute of Technology have developed particular AI algorithms – called “liquid” networks – capable of continually adapting to new data inputs. An unprecedented artificial neural network, which could help in the decision-making process applied to diagnostic imaging and autonomous driving.
Artificial neural networks are computational models composed of artificial neurons that emulate – simplifying it – the biological neural network. Their areas of application include, among others, vehicle control, process control, face identification, object recognition and diagnostics.
Researchers at MIT – Massachusetts Institute of Technology have developed a particular type of artificial neural network, capable of learning while it works and not only during its training phase.
In detail, these are flexible AI algorithms – called “liquid” networks – capable of modifying the underlying equations to continuously adapt to new data inputs. The outcome of this research – which will be presented at the AAAI conference, from 2 to 9 February 2021, on artificial intelligence – could give an important boost to the work on decision-making processes based on data flows that change over time, including those involved in diagnostic imaging and autonomous driving.
Flexible artificial neural networks
According to Hasani, time series data is ubiquitous and vital to our understanding of the world. The real world is all about data sequences. And also our perception. In reality – explains the professor – we do not perceive images, but “sequences of images”.
The vicissitudes of ever-changing data streams can be unpredictable. But analyzing this data in real time and using it to anticipate future behavior may accelerate some areas of application, including, for example, autonomous driving.
Following this thesis, Hasani’s team designed an artificial neural network capable of adapting to the variability of real-world systems.
Artificial neural networks are algorithms that recognize patterns by analyzing a series of examples of “training”. And they are able to mimic the processing mechanisms of the brain. For this, the group of seekers decided to take inspiration from Caenorhabditis elegans, a worm about 1 mm long, which lives in the soil: it has only 302 neurons in its nervous system – they explain – and yet it is capable of generating unexpectedly complex dynamics.
The advantages of liquid neural networks
But there is another advantage offered by the flexibility of the liquid network: it bypasses the inscrutability, common to other artificial neural networks, emphasizes the author of the research.
Only by changing the representation of a neuron (which Hasani did with differential equations) can one explore certain degrees of complexity that otherwise would not be possible.
Thanks to the small number of highly expressive neurons used by the team, it is easier to be able to scrutinize the “black box” of the network’s decision-making process and diagnose why it has carried out a certain characterization.
The model itself is richer in terms of expressiveness. And this could, in the future, help engineers increasingly understand – and improve – the performance of the liquid network.
The network model developed by the MIT researchers stood out in a number of tests. More specifically, it outperformed other state-of-the-art time sequence algorithms by a few percentage points, especially in accurately predicting future values in datasets for applications ranging from atmospheric chemistry to traffic models.