Explore ensemble techniques

[kyanmahajan]

Ensemble Models Using ResNet Embeddings

Overview

Ensembling is a machine learning technique where predictions from multiple models are combined to achieve better performance, robustness, and generalization than a single model. By leveraging complementary strengths of different models, ensembles often reduce variance and improve prediction stability.

In this task, you will build an ensemble using ResNet-based embeddings and evaluate its performance against individual models.

---

Task Description

1. Use a pretrained ResNet model to extract embeddings from images.
2. Train 2–3 different models on top of these embeddings (for example: Logistic Regression, SVM, MLP, Gradient Boosting, etc.).
3. Ensemble the trained models using one of the following approaches:
   • Averaging predictions
   • Weighted averaging
   • Voting (hard or soft)
4. Compare the ensemble performance with each individual model.
5. Briefly document your approach and observations.

---

Folder Structure (Important)

All work **must be done inside your own subfolder in participants folder ** following this structure: