Projects

Here, I have listed some of my works, based on their applications. Please click on each topic to go the main project idea and its implementation.

• Brain Decoding

  ▼

  • Brain Decoding Using Connectivity Informed Models

    Main idea Code Jupyter Book Slides

    Brain decoding is a procedure for descovering the brain activity, based on its encoded data. Here a graph neural network approach has been used for training, and task classification of encoded data.

• Image Denoising

  ▼

  • Image Multiplicative Noise Removal Using Statistical Modeling In The Shearlet Domain. (M.Sc. Thesis)

    Main idea Code

    Noise, and distortions, especially multiplicative noise, cause granular effects on images. The pivotal point of this work is to remove such effects by transferring the image to the shearlet domain, and statistically model the coefficients to obtain an efficient estimator. For more details please check my publications. Matlab

    For implementations please check my publications.

• Time Series/ Sequence Analysis

  ▼

  • Forward Forward algorithm on speaker identification.

    Main idea Code

    Here I have implemented the forward-forward algorithm for the time series (speech) data, for my conversational AI course. The .ipynb file is self contained and has all the detailed information. I also used this implementation for digit classification and was extended to time series data. Python, Pytorch

  • Automatic Speech Recognition(ASR) using hybryd attention + CTC model.

    Main idea Code

    Using CRDNN model and CTC loss for ASR task. Python, speechbrain

  • Generative Language Model

    Main idea Code

    Here a transformer-based model is used to learn the distribution of text data and finally generate related data, as samples. Python, speechbrain

  • Machine Translation

    Main idea Code

    Machine traslation task is implemented using a seq2seq model and a GRU based Encoder/ Decoder Attentive model. Python, speechbrain

  • Air pollution prediction based on statistical models, random processes (GARCH / ARIMA(X) / SARIMA / ETS)

    Main idea Code

    Most of the time series are non-stationary, making it hard to follow and predict their future values. In this work, random processes mentioned above were used to learn some patterns like trends, seasonality, volatility clustering, etc. Having this gives us the flexibility to tackle with small amount of data and easier prediction on time series. Python, Stats.model.

    This is a private project for AUT and the code is not shareable.

• Watermarking

  ▼

  • Statistical Analysis of Contourlet Coefficients of Watermarked Images.

    Main idea Code

    Designing an efficient statistical detector to classify the watermarked coefficients from the clear one. The Generalized Gaussian as a prior for contourlet coefficient is employed, and ROC curve is used to evaluate the designed statistical detector. Matlab

  • Injecting a text message into a 12 channels electrocardiogram signal based on pseudo-random permutation. (B.Sc. Project)

    Main idea Code

    The key point relies on erratically, by a pseudo-random permutation algorithm, change the channels order, without information loss. Matlab

• Big Data Analysis

  ▼

  • High dimentional KNN algorithm with CUR matrix approximation

    Main idea Code

    Implementation of KNN algorithm based on a dimension reduction algorithm (CUR decomposition as low rank approximation of SVD). Python, Matlab

  • High dimentional KNN algorithm with Locality Sensitive Hashing(LSH)

    Main idea Code

    Implementation of KNN algorithm based on Locality Sensitive Hashing(LSH). Python