Approximating probabilistic models as weighted finite automata

Weighted finite automata (WFA) are often used to represent probabilistic models, such as n-gram language models, since, among other things, they are efficient for recognition tasks in time and space. The probabilistic source to be represented as a WFA, however, may come in many forms. Given a generic probabilistic model over sequences, we propose an algorithm to approximate it as a weighted finite automaton such that the Kullback-Leibler divergence between the source model and the WFA target model is minimized. The proposed algorithm involves a counting step and a difference of convex optimization step, both of which can be performed efficiently.We demonstrate the usefulness of our approach on various tasks, including distilling n-gram models from neural models, building compact language models, and building open-vocabulary character models. The algorithms used for these experiments are available in an open-source software library.

Fractal Coding Based Video Compression Using Weighted Finite Automata

Main objective of the proposed work is to develop an approach for video coding based on Fractal coding using the weighted finite automata (WFA). The proposed work only focuses on reducing the encoding time as this is the basic limitation why the Fractal coding not becomes the practical reality. WFA is used for the coding as it behaves like the Fractal Coding (FC). WFA represents an image based on the idea of fractal that the image has self-similarity in itself. The plane WFA (applied on every frame), and Plane FC (applied on every frame) coding approaches are compared with each other. The experimentations are carried out on the standard uncompressed video databases, namely, Traffic, Paris, Bus, Akiyo, Mobile, Suzie etc. and on the recorded video, namely, Geometry and Circle. Developed approaches are compared on the basis of performance evaluation parameters, namely, encoding time, decoding time, compression ratio, compression percentage, bits per pixel and Peak Signal to Noise Ratio (PSNR). Though the initial number of states is 256 for every frame of all the types of videos, but we got the different number of states for different frames and it is quite obvious due to minimality of constructed WFA for respective frame. Based on the obtained results, it is observed that the number of states is more in videos namely, Traffic, Bus, Paris, Mobile, and Akiyo, therefore the reconstructed video quality is good in comparison with other videos namely, Circle, Suzie, and Geometry.

Human Activity Recognition through Weighted Finite Automata

This work addresses the problem of human activity identification in an ubiquitous environment, where data is collected from a wide variety of sources. In our approach, after filtering noisy sensor entries, we learn user’s behavioral patterns and activities’ sensor patterns through the construction of weighted finite automata and regular expressions respectively, and infer the inhabitant’s position for each activity through frequency distribution of floor sensor data. Finally, we analyze the prediction results of this strategy, which obtains 90.65% accuracy for the test data.

Fractal Coding Based Video Compression Using Weighted Finite Automata

Main objective of the proposed work is to develop an approach for video coding based on Fractal coding using the weighted finite automata (WFA). The proposed work only focuses on reducing the encoding time as this is the basic limitation why the Fractal coding not becomes the practical reality. WFA is used for the coding as it behaves like the Fractal Coding (FC). WFA represents an image based on the idea of fractal that the image has self-similarity in itself. The plane WFA (applied on every frame), and Plane FC (applied on every frame) coding approaches are compared with each other. The experimentations are carried out on the standard uncompressed video databases, namely, Traffic, Paris, Bus, Akiyo, Mobile, Suzie etc. and on the recorded video, namely, Geometry and Circle. Developed approaches are compared on the basis of performance evaluation parameters, namely, encoding time, decoding time, compression ratio, compression percentage, bits per pixel and Peak Signal to Noise Ratio (PSNR). Though the initial number of states is 256 for every frame of all the types of videos, but we got the different number of states for different frames and it is quite obvious due to minimality of constructed WFA for respective frame. Based on the obtained results, it is observed that the number of states is more in videos namely, Traffic, Bus, Paris, Mobile, and Akiyo, therefore the reconstructed video quality is good in comparison with other videos namely, Circle, Suzie, and Geometry.