Ph.D. Candidate in Information Systems, UMBC
Blog Topic
Articles on machine learning, deep learning, forecasting, decision trees, graph neural networks, and knowledge distillation.
A clear look at Milan's UAM demand forecasting study, including airport shuttles, intercity services, air taxis, vertiport scenarios, fares, and traveler preferences.
A practical explanation of how neural networks can accelerate genetic algorithms for pre-disaster air mobility evacuation planning.
A clear explanation of stochastic OD matrix forecasting with matrix factorization, graph convolutional networks, and recurrent neural networks.
A readable walkthrough of a four-step UAM demand forecasting model for Chengdu, China, with a focus on modal split, shared operations, and 2030 demand.
A clear guide to symbolic knowledge distillation for LLMs, including direct distillation, multilevel distillation, reinforcement learning, and why symbolic outputs matter.
A readable timeline of how language models evolved from early symbolic and statistical systems to transformers, LLMs, and symbolic knowledge distillation.
A clear guide to knowledge distillation, including response-based, feature-based, relation-based, and symbolic knowledge distillation.
A practical look at how machine learning can classify air taxi demand across locations and time periods using ride and weather data.
A readable explanation of using support vector regression to forecast final airline bookings from early pre-sale booking patterns.
A readable explanation of using LSTM, GRU, and Transformer models to estimate future Urban Air Mobility demand from taxi-trip data.
A practical explanation of using Temporal Fusion Transformers to improve multi-horizon airport departure demand forecasting.
A readable explanation of using seq2seq and attention-based deep learning models to forecast airport departure demand for strategic planning.
A readable explanation of a hybrid radial basis function network and decision tree approach for supporting breast cancer prognosis.
A practical explanation of how symbolic equations can be extracted from trained deep learning models by using inductive biases, sparse latent spaces, graph networks, and symbolic regression.
A practical introduction to graph neural networks, message passing, graph network blocks, and relational inductive bias for learning over structured data.
A practical explanation of Budding Trees, a differentiable decision-tree model where each node can smoothly move between being a leaf and an internal split.
A practical explanation of Adaptive Neural Trees, a model that combines the representation learning power of neural networks with the conditional structure and lightweight inference of decision trees.
A practical explanation of soft decision trees, where internal nodes route examples probabilistically so every leaf can contribute to the final prediction.
A practical explanation of Frosst and Hinton's method for distilling a neural network into a soft decision tree, trading some accuracy for a more interpretable model.
A practical overview of the major challenges in deep reinforcement learning, including sample inefficiency, poor generalization, interpretability, reproducibility, safety, and real-world deployment.
A clear explanation of Informer, an efficient Transformer architecture for long sequence time-series forecasting that uses ProbSparse attention, self-attention distilling, and a generative decoder.