Page Not Found
Page not found. Your pixels are in another canvas.
Sitemap
A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.
Pages
Page not in menu
This is a page not in th emain menu
Posts
Variational Autoencoders
Published:
Variational Autoencoders (VAEs) represent a fundamental breakthrough in generative modeling, combining the power of deep neural networks with principled Bayesian inference. Introduced by Kingma and Welling (2013) [1], VAEs provide a scalable framework for learning complex probabilistic models with continuous latent variables.
Introduction to Generative Modeling
Published:
This blog provides a primer on generative modeling, summarizing introductory knowledge required before venturing into more advanced topics. It will serve as a foundational piece that I plan to build upon and update whenever necessary. As I delve deeper into the subject, I intend to create more comprehensive articles focusing on various topics, This blog acts as an initial stepping stone towards that goal.
Mathematical Derivation and Python Implementation of Principal Component Analysis
Published:
Dimensionality reduction is the transformation of data from a higher dimensional space to low-dimensional space, such that the information loss is minimum. Dimensionality reduction can be used for noise reduction, data visualization, cluster analysis, or as an intermediate step to facilitate other analyses. Principal Component analysis(PCA) is one such technique.
publications
eptk: energy prediction toolkit
Published in Proceedings of the 9th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, 2022
Conference Core Rating : A
Building energy use prediction plays a crucial role in whole building energy management. In recent years, with the advent of advanced metering infrastructures that generate sub-hourly energy meter readings, data-driven energy prediction models have been implemented by leveraging advanced machine learning algorithms. However, the lack of standardization of model development and evaluation tools hinders the advancement and proliferation of data-driven energy prediction techniques on a large scale. This paper presents eptk, an open-source toolkit that enables the seamless development of data-driven energy prediction models. The proposed toolkit helps researchers and practitioners to easily benchmark the existing and new data-driven models on various open-source datasets containing time-series of multiple energy meter data along with relevant metadata. Using the toolkit, we develop and compare the performance of 34 models on two large datasets containing more than 3,000 smart meter readings. eptk will be released in open-source for community use.
Download Paper
Interpretation of Drop Size Predictions from a Random Forest Model Using Local Interpretable Model-Agnostic Explanations (LIME) in a Rotating Disc Contactor
Published in Industrial & Engineering Chemistry Research, 2023
Drop size is a crucial parameter for the efficient design and operation of the rotating disc contactor (RDC) in liquid–liquid extraction. The current work focuses on providing local and global explanations for the prediction of the drop size in a rotating disc contactor (RDC). The Random Forest (RF) regression model is a robust machine learning algorithm that can accurately capture complex relationships in the data. However, the interpretability of the model is limited. In order to address the issue of interpretability of the developed RF model, in the current work, we employed Local Interpretable Model-Agnostic Explanations (LIME) of the predictions of the RF model. This provides both local and global views of the model and thereby helps one to gain insights into the factors influencing predictions. We have provided local explanations depicting the impact of different attributes on the prediction of the output for any given input example. We have also obtained global feature importance, providing the top subset of informative attributes. We have also developed local surrogate models incorporating second order attribute interactions. This has provided important information about the effect of interactions on the drop size prediction. By augmenting the random forest model with LIME, it is possible to develop a more accurate and interpretable model for estimating the drop size in RDCs, ultimately leading to improved performance and efficiency.
Download Paper
Explainable AI for Energy Prediction and Anomaly Detection in Smart Energy Buildings
Published in Proceedings of the 10th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation, 2023
Conference Core Rating : A
Buildings consume significant energy, accounting for approximately 40% of total energy usage worldwide. Unfortunately, on average, commercial buildings waste around 30% of the energy they consume. In recent years, the advancement of artificial intelligence and smart metering infrastructure has led to the emergence of datadriven methods for energy prediction and anomaly detection. These methods provide automated decision support to building operators in managing and preventing energy loss. Despite the advantages of having sophisticated data-driven models, one major drawback is their lack of transparency. This paper focuses on enhancing the transparency of data-driven models for energy prediction and anomaly detection. The research investigates the utilization of SHapely Additive exPlanations (SHAP), an explainable AI algorithm, to provide insights into large-scale energy prediction and anomaly detection models. Additionally, the present study introduces a framework that seamlessly integrates feature transformations within the model, while SHAP operates on the interpreatable feature space, enhancing the explanations provided by SHAP values.
Download Paper
Generative Adversarial Network with Soft-Dynamic TimeWarping and Parallel Reconstruction for Energy Time Series Anomaly Detection
Published in Proceedings of AI4TS workshop of AAAI 24, 2024
Conference Core Rating : A*
In this paper, we employ a 1D deep convolutional generative adversarial network (DCGAN) for sequential anomaly detection in energy time series data. Anomaly detection involves gradient descent to reconstruct energy sub-sequences, identifying the noise vector that closely generates them through the generator network. Soft-DTW is used as a differentiable alternative for the reconstruction loss and is found to be superior to Euclidean distance. Combining reconstruction loss and the latent space’s prior probability distribution serves as the anomaly score. Our novel method accelerates detection by parallel computation of reconstruction of multiple points and shows promise in identifying anomalous energy consumption in buildings, as evidenced by performing experiments on hourly energy time series from 15 buildings.
Download Paper
Protein Feature Engineering Framework for AMPylation Site Prediction
Published in Nature Scientific Reports, 2024
AMPylation is a biologically significant yet understudied post-translational modification where an adenosine monophosphate (AMP) group is added Tyrosine and Threonine residues primarily. While recent work has illuminated the prevalence and functional impacts of AMPylation, experimental identification of AMPylation sites remains challenging. Computational prediction techniques provide a faster alternative approach. The predictive performance of machine learning models is highly dependent on the features used to represent the raw amino acid sequences. In this work, we introduce a novel feature extraction pipeline to encode the key properties relevant to AMPylation site prediction. We utilize a recently published dataset of curated AMPylation sites to develop our feature generation framework. We demonstrate the utility of our extracted features by training various machine learning classifiers, on various numerical representations of the raw sequences extracted with the help of our framework. 10-fold cross-validation is used to evaluate the model’s capability to distinguish between AMPylated and non-AMPylated sites. The top-performing set of features extracted achieved MCC score of 0.58, Accuracy of 0.8, AUC-ROC of 0.85 and F1 score of 0.73. Further, we elucidate the behaviour of the model on the set of features consisting of monogram and bigram counts for various representations using SHAP (SHapley Additive exPlanations.
Metaheuristic and Evolutionary Algorithms in Explainable Artificial Intelligence
Published in Springer, part of the book series: Computational Intelligence Methods and Applications, 2024
There has been an increasing interest in Explainable Artificial Intelligence (XAI) in recent years. Complex machine learning algorithms, such as deep neural networks, can accurately predict outcomes, but provide little insight into how the decision was made or what factors influenced the outcome. This lack of transparency can be a major issue in high-stakes decision-making scenarios, where understanding the reasoning behind a decision is crucial. XAI aims to address the problem of the ”black box” in machine learning models, where the AI’s decision-making process is not transparent, and humans cannot understand how the AI arrived at a particular decision or prediction. Evolutionary and metaheuristic techniques offer promising avenues for achieving explainability in AI systems, and there is a lot of ongoing research in this area to further explore their potential. Our work is a concise literature review that explores the potential adoption of these techniques to facilitate the attainment of explainability in AI systems.We have highlighted some of the contributions of evolutionary and metaheuristic techniques in different approaches to achieving explainability, such as counterfactual explanations, local surrogate modelling, and the development of transparent models.
Download Paper