Human Activity Recognition
Human Activity Recognition
Embedded Files
Built a model that predicts human activities such as Walking, Walking Upstairs, Walking Downstairs, Standing, Sitting, or lying using sensors like an accelerometer to measure acceleration and a gyroscope to measure angular velocity that we have in the Smartphone.
The dataset is collected from 30 persons performing different activities with a smartphone to their waist and it is recorded from the tri-axial reading of the accelerometer and gyroscope measured w.r.t time. So, we have 6 time series data as input to solve a 6-class multi-class classification problem.
The sensor signals (accelerometer and gyroscope) were pre-processed by applying noise filters and then sampled in fixed-width sliding windows of 2.56 sec and 50% overlap (128 readings/window).
We have 561 hand-coded features which are engineered by the domain experts and obtained as a frequency and amplitide variation of time series data.
Apart from that, we have raw time series collected directly from the sensors which are then fed into the deep learning Algorithms to auto-engineer all the features and make the predictions.
Hyper-tuned all the relevant machine learning models on the Handcoded 561 features for the human activity recognition problem
ML Algorithms used: ML - Logistic Regression, Linear SVC, SVM, Decision Tree, Random Forest.
DL Algorithms used - Long Short-Term Memory
Project URL: Humanella website, YouTube link
Music Generation Using AI and Big Data Tools
Music Generation Using AI and Big Data Tools
This project aims to leverage artificial intelligence (AI) techniques to generate high-quality music compositions using the MusicNet dataset and Big Data Tools
We analyzed the MusicNet dataset, which provided valuable information about the distribution of ensembles and compositions across different composers. Solo piano emerged as the most common ensemble type, while Beethoven stood out as the dominant composer.
We leveraged the power of LSTM and GAN models for music generation, training them using GCP services such as GCP Bucket for data storage, GCP Dataproc, and Pyspark for distributed processing, GCP Notebooks for development, and FastAPI for deployment.
We employed various callbacks in the training process, including learning rate scheduling, reducing learning rate on plateau, model checkpointing, and TensorBoard visualization. These callbacks played crucial roles in controlling the training process and monitoring model performance.
While we initially experimented with Elephas(Pyspark Library), an integration of Keras and Spark, we ultimately switched to TensorFlow for its flexibility, compatibility, and extensive community support.
Overall, our journey through music generation using deep learning and GCP showcased the potential of these technologies in creating AI-powered music compositions. By combining cutting-edge machine learning algorithms, cloud computing resources, and efficient development and deployment tools, we unlocked new possibilities in music creation and expanded our understanding of the intersection between AI and music.
FastAPI is a Python framework for building APIs. Combined with GCP, we tried deploying our LSTM models.
We can explore advanced AI techniques such as reinforcement learning and generative adversarial networks to further enhance the quality and diversity of the generated music compositions. Additionally, we can consider incorporating additional features and metadata into the models to capture more nuanced musical characteristics.
Incorporating real-time streaming data, allowing for dynamic and interactive music generation. This would involve implementing scalable and distributed streaming data processing frameworks such as Apache Kafka and Apache Flink.
Another interesting direction is to explore the application of transfer learning in music generation. By leveraging pre-trained models on large music corpora, we can transfer the learned knowledge and patterns to improve the performance and efficiency of our models.
Image Caption Generation using PySpark
Image Caption Generation using PySpark
Objective:
Whenever an image appears in front of us, our brain can annotate or label it. But what about computers? How can a machine process and label an image with a highly relevant and accurate caption? It seemed quite impossible a few years back. In this project, I built one such annotation tool capable of generating relevant captions for the image with the help of AI and ML
Whenever an image appears in front of us, our brain can annotate or label it. But what about computers? How can a machine process and label an image with a highly relevant and accurate caption? It seemed quite impossible a few years back. In this project, I built one such annotation tool capable of generating relevant captions for the image with the help of AI and ML
Algorithms used:
DL - Xception (Feature Extraction), Long Short-Term Memory (Caption Generation)
Project URL: GitHub link
Analysis of Public Sentiments on Climate Change for Formulating Policies
Analysis of Public Sentiments on Climate Change for Formulating Policies
Objective:
• Addressed the research question, ”To what extent can public sentiments on social media platforms such as Twitter influence national or global policies on climate change?”
• Addressed the research question, ”To what extent can public sentiments on social media platforms such as Twitter influence national or global policies on climate change?”
• Employed Named Entity Extraction techniques to identify relevant organizations or locations within the discourse.
• Employed Named Entity Extraction techniques to identify relevant organizations or locations within the discourse.
• Observed that the Logistic Regression model achieved the highest accuracy of approximately 70% compared to other models, providing valuable insights into the research question’s context and potential implications for policy formulation.
• Observed that the Logistic Regression model achieved the highest accuracy of approximately 70% compared to other models, providing valuable insights into the research question’s context and potential implications for policy formulation.
Algorithms used:
ML - Logistic Regression, Linear SVC, SVM, Decision Tree, Random Forest
Project URL: GitHub link
Page updated
Report abuse