CertiLand : Shielded RL with Temporal-Logic Verification
CertiLand : Shielded RL with Temporal-Logic Verification
Shielded Reinforcement Learning with Temporal-Logic Verification for the safe landing of unmanned aerial vehicles (UAVs) during exploration is a crash, a no-fly-zone incursion or a battery depletion in the air. CertiLand project presents a framework that makes model-free RL for UAV landing safe during learning and certifiably safe after it, without control barrier functions, without demonstration data, and without slack variables. A battery- and deadline-aware safety requirement is written in linear temporal logic (LTL) and compiled by formula progression into a deterministic monitor. A safety game on the product of the UAV Markov decision process and the monitor, in which wind gusts act adversarially, yields a shield that removes exactly those actions that could violate the specification under some gust sequence. We prove that every run of any shield-compliant learner satisfies the specification, that the shield is maximally permissive, and that a relaxed probabilistic shield keeps the safety probability by an independent certificate. After learning, the deployed policy is certified by exact model checking: computation tree logic (CTL) on the closed-loop Kripke structure for the worst case, and probabilistic CTL (PCTL) on the closed-loop Markov chain for the average case.
This project, Heart-diseases-Prediction, is a comprehensive study of heart disease prediction using various machine learning algorithms such as Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Decision Trees, and Random Forests. The project employs a dataset of heart disease cases to train and test these models, aiming to predict the likelihood of heart disease based on various patient parameters. The results obtained from these models can potentially aid healthcare professionals in early detection and treatment planning for heart diseases.
cite our work on Google Scholar
Data-Driven Analysis and Visualisation of Cricket Player Statistics
This project Data-Driven Analysis and Visualization of Cricket Player Statistics', is an extensive exploration of a dataset containing comprehensive information about cricket players. The project involves loading, cleaning, and analyzing the data, then visualizing the results using various tools and libraries. The dataset includes details such as the player's name, span, matches, innings, not outs, runs, hundreds, average, fours, sixes, centuries, fifties, zeros, fours, and sixes.
The project also includes a detailed exploratory data analysis (EDA), which includes checking for missing values, visualizing the data, and performing transformations on the data. The analysis is conducted using Python libraries such as Pandas, Numpy, Matplotlib, and Plotly. The project also involves creating visualizations like bar charts, scatter plots, and histograms to better understand the data.
This project aims to detect depression from EEG signals using Python libraries such as MNE and NumPy. The project involves several steps:
1. Data Collection: The EEG data is collected from a Figshare repository using the `requests` library. The data is then extracted and saved in a local directory.
2. Data Preprocessing: The raw EEG data is preprocessed using the MNE library. This involves resampling the data to a target sample rate, applying band-pass filtering, and saving the preprocessed data.
3. Data Visualization: The preprocessed data is then visualized using MNE's built-in plotting functionality. This provides a visual representation of the EEG signals, which can be useful for further analysis.
4. Feature Extraction: The preprocessed data can then be used for feature extraction. This step is not shown in the provided code but is an important part of the process. Features extracted from the EEG signals can be used to build machine-learning models for depression detection.
the data is loaded from the EDF files, resampled to 250 Hz, and then band-pass filtered between 0.5 Hz and 40 Hz. The preprocessed data is then saved in a NumPy array format for further use.
The preprocessing steps are essential to ensure that the EEG signals are in a suitable format for further analysis. Resampling ensures that the data has a consistent sample rate, while band-pass filtering removes noise and other irrelevant frequency components.
IEEE Access [1] Smart City Traffic Management: Acoustic-Based Vehicle Detection Using Stacking-Based Ensemble Deep Learning Approach. (2024)
JCBI [2] Exploiting Machine Learning Models for Identification of Heart Diseases. (2022)
IEEE Conference [3] Ensemble-Based Approach for Heart Disease Prediction. (2022)
JCBI [4] Prediction of Electric Power Demand of HVACs for Operating Rooms in Case of Dynamic Set Points of Temperature: A Case Study, (2022) World scientific
world scientific [5] Clinical Decision Support System (CDSS) for heart disease diagnosis and prediction by Machine learning Algorithms: A systematic literature review. (2023)
JCBI [6] Shabbir, A., & I. Ullah, K. (2023). Brain Tumor Detection Based on Deep Learning Approach. Journal of Computing & Biomedical Informatics, 4(02), 298-310.