Embedded Files
The "Geometric Deep Learning and Graph Machine Learning" research pod is dedicated to advancing and applying state-of-the-art techniques in graph-based learning and geometric data representation. With the rising importance of structured and relational data across multiple domains, this pod focuses on leveraging graph neural networks (GNNs), topological data analysis, and geometric learning frameworks to solve real-world problems.
In addition to original research contributions, this pod fosters knowledge sharing through an active Reading Group that delves into seminal papers, recent breakthroughs, and interdisciplinary applications of geometric learning.
Core Research Focus
Differential Geometry
Foundation Models
Geom. Diffusion Models
Algorithmic Reasoning
In recent years there is an exponential growth of research in the field of recommendation systems(RS), with the evolution of graph machine learning to learn a better representation of interactions between entities Because the data on customer behavior is observational instead of experimental, it contains many biases. If the data are simply fitted without taking these biases into account, major problems would arise. However, most researchers try to model the recommendation task by the means of graph-based algorithms and characterize it better than collaborative filtering-based approaches. Still, the problem of recommendation fairness challenges the architectures, and the issue of biases in the graph-based recommendation system gets un-explored. In this work, we first formulate recommendations as link predictions between users and items in a temporally evolving graph to update the user’s preferences over time, and then we establish the effect of exposure bias in the recommendations, Finally, we propose some computationally in-expensive methods to reduce the effect of bias, along with an end-to-end differentiable framework (“diffDebias”) to redistribute the biased data distribution to an ideal one. Empirical results show that by adopting a bias-fairness tradeoff we model a more fair recommendation system than the most accurate one.
In this work we propose a three-step process to generate the graph in an unsupervised manner, the nodes and topological information is captured and embedded using Node2Vec based biased random walks and these node embeddings are used to generate edges using a variational autoencoder based model, and the latent representation is used to connect node embedding via learning the explicit representation of the network, now the generation of the graph is done in a sequential manner using recurrent neural networks with the help of previously obtained embedding of nodes and edges, in this manner the generated graph contains similar properties of the network but different structural information, the resultant graph exhibit temporal evolution of the existing graph. The empirical analysis of the results obtained graph depicts the 90% structural similarity between graphs, achieving state-of-the-art results on the evolution models.
Resources and Activities
If you are a new comer in the field of graph machine learning and want to pursure the challenging problems, CS224W : Machine Learning with Graphs is a great place to start.
Google Sites
Report abuse