Altair Global Contest for Students - 2022 - FAQ
Altair EmployeeFrequently Asked Questions regarding Altair Global Contest for Students - 2022 :
Can someone participate in a team?
__Yes, you can participate as a team of students from your university or across universities.
Would we have any sort of live training webinar and when?
__ A live webinar was conducted and the recording is put up on community registration page of the contest and the contest webpage, Further Webinars are scheduled and the joining information and dates will be announced in the community and the contest page - https://altairuniversity.com/contest/
Do participants require any prior experience or knowledge before participating?
__No, the training materials and webinar on the contest webpage start from the very basics.
How to acquire test data from a real world model?
__ Data can be acquired from real world systems by means of a datalogger and appropriate sensors.
What are the minimum machine specifications should I have to run the simulations smoothly?
__Student laptops and PCs are sufficient to run the contest problem
Do I have to download ROM AI from the student license mail as well or just Activate?
__romAI is preinstalled in Altair Activate 2022, just go to File > Extension Manager > Switch it on.
How do I open the H3D file?
__ H3D files are a results-container file. You may use HyperView Player available through the installation of the HyperWorks EDU version to open H3D files. Alternately, they can be viewed in HyperView if the node count is within the limit set by the student license.
(More information about student version limitations at https://altairuniversity.com/altair-student-edition-limitations/)
Where do I find the hwdesktop and hwsolvers files to set the Activate paths?
__ Please refer to this solved Altair Community question: https://community.altair.com/community?id=community_question&sys_id=63fdc4dddb766410cfd5f6a4e2961967
What is a neural network architecture?
__Net Architecture refers to the structure of the Neural Network(NN) in terms of the number of neurons in a layer (width) and the number of hidden layers (depth) in the network.
A network will have an input and output layer, and these layers will have a number of neurons equal to the number of inputs and outputs provided for in the data. The number of neurons per layer and the number of layers must then be selected to provide the network with enough flexibility to “learn” to map the inputs and predict the outputs from provided data.
What is Net Architecture and how do we choose the parameters such as Neurons and Hidden layers?
__ These parameters, and a few more are collectively also known as Hyperparameters, ie parameters that determine a Neural Network’s structure.
The problem of the choice of hyperparameter values is a challenging one.
Look at it this way - You are doing a trial and error approach by changing two hyperparameters (# of HL, # of NN) and observe the trend of how much the AI model deviates from your training data
You can manually begin to look for a reasonable network architecture for your specific scenario by using trial values for the number of Hidden Layers and the number of Neurons per hidden layer and observing the trend of the magnitude of overall loss for a relatively small number of epochs. The ideal is a rapidly decreasing value of loss.
In general, adding more neurons to a layer tends to increase the NN’s ability to reproduce the original system (fitting capability), but adding too many risks overfitting, i.e. the ability to reproduce known responses to known inputs but poor ability to predict outputs with previously unseen inputs.
And, in general, adding more hidden layers has the effect of allowing the NN to smoothly approximate outputs, whilst allowing for more conformance of the NN to the real world system.
What is an epoch?
__Simply put, an epoch count or the number of epochs is a number that indicates how many times the neural network has seen all of the training examples, i.e. when all the data points set aside for training have been presented to the network an epoch is said to have passed (the epoch count increases by 1). In general, the larger the number of epochs that you specify that a network must train for the better it may “learn” the system represented by the data.
How do I decide the number of epochs?
__The romAI Director tool will helpfully suggest a minimum number of epochs based on criteria like the number of Input/Output/State variables and the neural network structure (ie. Number of neurons per layer and the number of layers). Do try a value that is larger than this minimum to see if training with the currently specified network architecture results in gradually decreasing loss.
The upper limit will be dependent on factors like the size of the dataset, the required accuracy (Although it is not true that running for a larger number of epochs will always result in better trained neural networks.)
How to plot HyperPlanes?
__ Hyperplanes are available only when you have Outputs that are not also state variables. In some cases, all of your Outputs can also be State variables, in such a case the HyperPlanes option is not available in the options in the Viewer tab of romAI Director.
In cases where your Outputs are different from the State variables, you may plot Hyperplanes for one output at a time.
You may either create a Prediction vs Target plot by providing a csv file with a similar data structure to that of the training data, or, you may clear the `Predict on Dataset` checkbox and provide your own input ranges for the selected Input variables.
Where do I find basic Activate tutorials?
__ With Activate started up go to File > Help > Tutorials… and you will be directed to the help website with the available tutorials.
Through your Altair One (community account) you can access the learn.altair.com e learning system with detailed and introductory Activate courses - https://learn.altair.com/totara/catalog/index.php?cfp_multiselect_categoriescateg_9d3e8[]=Systems Modeling&catalog_fts=activate&orderbykey=score&itemstyle=narrow
Have a look at the learning library activate material- https://altairuniversity.com/learning-library/?type=learninglibraryitem&filter_discipline=Activate&search&filter_resource_type&filter_language&filter_source
FREE eBook: Learn Basics of System Modeling and Control Systems with Altair Activate - https://altairuniversity.com/free-ebook-system-modeling-activate/
Courses in the learning and certification system - https://certification.altairuniversity.com/course/index.php?categoryid=54
Answers
-
Any questions or concerns regarding the contest can be asked in the competitions forum here and will be answered promptly.
0