Praveen Kumar

- Analyse and Visualization of data with multiple types of graphs.
- After doing appropriate data preprocessing we use dimensionality reduction technique NLPCA for visualization of categorical association between attributes

- Implemented a task-dependent SalGAN network in Tensorflow and task-dependent SAlNet network in Keras using custom loss function with Salicon dataset.
- Compare the result with different type of saliency approaches.
- One can see our results output on my Youtube channel.

- We formed and visualize Concept maps, aka "Knowledge Graph" between topics and their relevant words from text dataset using Natural Language Processing.
- Code and result on my GitHub repository.

- Implemented Bidirectional LSTM encoder of input sentences and max-pooling operation over timesteps.
- A mathematical technique was used to fetch information to produce a relationship between inputs and then classify.

- Built a security system with the help of microcontroller ATmega2560. The system includes tracking location in a given range, alarm alert message and sound, operated by mobile app and password protected which is stored in SD card.
- Code and other detail of project can be found on my GitHub repository.

- Built an Obstacle avoiding Robocar using Atmenga320p, Proximity sensor, IR, HC-05 Bluetooth module, servo and DC motors.
- Code and other detail of project can be found on our GitHub repository.

- Research on similarity detection using machine learning and image processing techniques.
- We have to find the simalrity betweeen two object images. Objects can be anything like refrigerator, tv, washing machine, ac etc. We develop a algorithm with fusion of image processing and a deep convo neural networks to detect the similarity.

- In this game 16 tiles are present in pairs (8 unique tiles). You have to find the indentical pairs of tile.
- You can find the description with code from here.

-Built a webpage for testing two cognitive tasks- Attention and Memory and launched the webpage using google cloud platform(GCP).
- Programming Language used for this task are HTML, CSS, Javascript, AngularJS.
- You can check this from http://35.238.141.106/ webpage and also can test your memory (beware from negative marking 😜)
- If link not open just copy and past in new tab.

- Built a Quadcopter using Arduino Mega as a flight controller in which we use brushless DC motors, battery with output voltage 11.1V, 2.4Ghz 6CH Transmitter and Receiver (approx. 2Km range in line of sight), I3G4250D 3-axis Gyroscope as components.

- In this project I built a classification model using CNN network that can classify crack and potholes with an accuracy of 94.85%.
- Used IP techniques like Hough transform and perspective transformation to find length and area covered by crack and potholes.

- In this project I built a solar mobile power bank with help of rechargeable battery cells, a solar plate, a buck converter that drop the voltage at 5V.
- We can charge the power bank with the help of both electricity and sunlight, this power bank has 12000mAH capacity.

- Forecast three electrical consumption meter reading with the help of different Data Analytics and Machine Learning techniques.
- Analyse and Visualize the dataset with different type of graphs, like bar graphs, heat maps, density plots, box plots, pie charts.
- Forecast data with ensemble different models(CNN-LSTM, FbProphet, Arima, LightGBM) after feature engineering on data.

- This application is developed for the class attendance and crowd monitoring. I built a model for face detection and recognition in video and count people in an image.
- Used OpenCV libraries and image processing techniques for face detection and recognition.
-Used CSRNet for counting people in an image, it uses VGG-16 in its starting layers.