Figure 3 - uploaded by Nikolay Manchev
Content may be subject to copyright.
Source publication
The aim of this project is to look into parallelization of Restricted Boltzmann Machines, their scalability and suitability for processing large data sets in the context of “big data”. The research question this project looks into is: Can a CPU-based, parallelized version of the Restricted Boltzmann Machine be developed and how does it compare to t...
Contexts in source publication
Context 1
... Boltzmann Machine comprises neurons that operate in binary fashion -they are either "on" (+1) or "off" (-1). The neurons are split into two functional groups: hidden and visible, as shown in Figure 3. ...
Context 2
... processing times (averaged over 10 epochs) for the individual feature files are shown in Table 11 and Figure 29. examples, the driver does not allocate and use resources in the executors (Figure 30a). The 20'000 examples dataset, however, requires a distributed computation as it does not fit in the driver's memory, so SystemML runs a series of distributed matrix operations that utilize Spark executor resources (Figure 30b). ...
Context 3
... the driver does not allocate and use resources in the executors (Figure 30a). The 20'000 examples dataset, however, requires a distributed computation as it does not fit in the driver's memory, so SystemML runs a series of distributed matrix operations that utilize Spark executor resources (Figure 30b). ...
Context 4
... plot of the residuals from the simple linear regression model (Figure 33) is also incon- clusive and may hint towards a curvilinear relationship. It is therefore suggested that more data would be needed to make a confident estimation about the type of the relationship and the scalability of the training process and the underlying infrastructure. ...
Context 5
... features it learns contain more noise compared to the features learnt by the DML RBM ( Figure 15 and Figure 17). While training the RBMs on the Gutenberg data, we also noticed that the decrease in the cost function is less clearly expressed and less stable in the Theano RBM in comparison to our custom RBM implementation ( Figure 34). ...
Similar publications
An experiment is performed to reconstruct an unknown photonic quantum state with a limited amount of copies. A semiquantum reinforcement learning approach is employed to adapt one qubit state, an “agent,” to an unknown quantum state, an “environment,” by successive single‐shot measurements and feedback, in order to achieve maximum overlap. The expe...
