Try out your first AM solution at $900 which includes 3 years of technical supports for a server computer instance, and a year of consulting and advisory.
The server comes with GPU options can cover a class of problems that can be handled with DNNs and relational CNNs. Upto the size 100000 of input vectors within a range of 12 hours training. Also a dedicated inference engine "IM" is provided that can handle as fast as 100 inferences a second if the size is some 100.
Notable details of the AI kernel are as follows.
᛫ training algorithms⠆ CD, BP, Softmax.
᛫ inference algorithms⠆ attended adaptive reconstruction,
x2y, y2x, various analytics, masked/focused inference.
᛫ parallelism⠆ batch based, loop based, tensor compute (GPU).
᛫ optimization⠆ simulated annealing and early stopping.
᛫ overfit handling⠆ dropout, edge down sampling, auto model refitting.
᛫ realtime handling⠆ time stampped file i/o.
᛫ depth handling⠆ inference caching for deep layers.
᛫ big data handling⠆ ssd based process for huge data size with options.
᛫ normalizations⠆ our own standardization method for input data,
layer standardization.
᛫ initializations⠆ Gaussian, HE Gaussian.
᛫ architectures⠆ deep feed forward, convolutional RBM, RBM, DBN.
There are many details that need to be attended to become a well formed solution. Developing such AI solutions of such quality requires a team of experts who are well aware of those issues, which must claim a decent amount of costs. The AM solutions will reduce the cost significanly because most relevant issues can be resolved with appropriate option settings without researching and coding that your business models need not pay attention to.
Our task is to deliver the ideas and expertise behind AM kernel solution as long as the kernel itself, to speed up your projects.