Welcome to my blog!
Posts
-
IRES Project: Dynamic Finetuning Of Multi-Task Multi-Modal Models
Over summer 2024, I got the chance to study at INAOE, a research university near Puebla, Mexico, in partnernship with University Of Houston as an IRES student. My partner and I worked on applying curriculum learning style techniques to finetune a Multi-Task Multi-Modal Model for Comic Mischief Detection. To learn more, see our final report here. In the report we discuss: the problem of Comic Mischief Detection; the multitask multimodal HICCAP model (pictured below); the difficulty of multiask learning; a survey of methods for training a multitask model; and our experimental results.
-
DBScan
DBScan is a useful density-based clustering algorithm for unsupervised learning. Unlike KMeans Clustering, DBScan can identify irregularly shaped clusters and is also resilient to noise. I go over the algorithm, its implementation, and two simple experiments. See the interactive Colab Notebook for more.
