Ashutosh Choudhary

Context based chatbot with a personality

our model

- We present an end-to-end deep RNN model to behave as a conversational agent having a particular persona. This con- versational system also remembers long range dependencies in the dialog with the help of attention mechanism and context based approach. A speaker model encodes personas in distributed embeddings which represent a variety of characteristics of the persona of the chatbot. These embeddings are learned from the data. Our model yields qualitative performance improvements over the base line sequence to sequence model with persona embeddings.

- Source Code

- Report

Music Recommendation using Recurrent Neural Networks

- A large amount of information is contained in the sequence of songs in a particular playlist. Typical collaborative filtering based recommendation approaches do not consider this information for rec- ommendations. This project models the music recommendation problem as a sequence prediction problem and explores the application of Long Short Term Memory (LSTM) networks to the problem.

- Source Code

- Report

Integrating OpenStreetMap with Google Street View using image processing and machine learning

Correct detection Incorrect detection

- Calculated the camera angle or heading of the Google Street View for each node in OpenStreetMap to show the correct picture of the node, out of the possible 360 s. A classifier trained on amenity label images evaluates the probability of camera angles on GSV images.

- Implemented fast sequential XML parser using LRU caching to parse large osm files upto 200GB

- Parallel computing to run classifiers simultaneously on several AWS EC2 clusters.

- Source Code

- Report

Adaptive Control of Traffic Grid Using Fuzzy Logic


Controller on nagpur map

An adaptive traffic controller using fuzzy logic is presented which takes into consideration the situation of its neighbors. The controller gives preference to less fuel consuming vehicles, public transport vehicles and high priority vehicles. A traffic grid system is designed using Simevents Toolbox of Simulink with provisions for turning and random class of vehicles. The results obtained from the implementation of this fuzzy logic controller are tabulated against those corresponding to a conventional cyclic fixed time controller, vehicle actuated controller and a basic fuzzy logic controller (designed for isolated junction). Simulations are done for varying traffic conditions namely heavy, medium, light and random. With the performance criterion being the average waiting time of vehicles, it is shown that the use of this fuzzy logic controller results in a better performance.

- Report