Nikolay Danielhin:

In this paper I describe a hybrid dynamic programming model for learning generative models using a mixed dynamic programming model. In contrast to generative methods which can be evaluated in different phases as the model evolves, my hybrid model is expected to evaluate the model using the same parameters in all phases, and therefore, when the model runs faster.

For the most part, dynamic programming models and generative methods follow the same underlying rules. The main difference is that in the latter case, the parameters that are evaluated evolve based on features found in the training, and not on the training parameters themselves. This change allows for efficient model generation without the need to compute hidden components.

I hope that this blog can be of interest to some of you. You can find some basic materials about how the model works in the previous posts I have written. If you have any questions about my model, feel free to comment below. This blog is not affiliated with Google.

Until next time!

Andreas Lehmann, M.A.

Prakash Kaul

Alexei J. Eremenko

Matheus M. van der Laan

Michai Kuchera

Hans-Uwe Nascimento

Andrey Zovko

Jürgen Tössner

David E. Yablonowski

David Eremenko:

This work was supported by the Deutsche Forschungsgemeinschaft, Deutsche Forschungswunder, Institut für Technik für Handelskunde, Institut für Bildung, Instituto de Ciencia y Ciencia Nacional de Desarrollo Científica, Universidad de Buenos Aires, University of Bologna, Universidad de Andalucía, Universidad de Las Palmas.