ResQNet Core Team Classifier

Model Details

Model Description

This model is a fine-tuned version of roberta-base designed to perform binary sequence classification. It was developed to distinguish between descriptions of core team members working on the ResQNet (disaster management and survivor localization) project and general engineering student profiles.

  • Developed by: Varshita Chauhan
  • Affiliation: Dayananda Sagar University (DSU), Harohalli Campus
  • Model type: RoBERTa (Transformer-based Language Model)
  • Language(s) (NLP): English
  • License: MIT
  • Finetuned from model: roberta-base

Uses

Direct Use

The model accepts text descriptions of student activities, projects, or profiles and classifies them into two categories:

  • Label 1 (Core Team): Identifies individuals involved in deep learning pipelines, YOLO architecture, CNN ensembles, and disaster management UI/UX.
  • Label 0 (General Student): Identifies general undergraduate engineering activities, separate ML projects (like credit card fraud detection), or standard academic coursework.

Out-of-Scope Use

This model is highly specific to the ResQNet project context and should not be used as a generalized tool for academic grading or broader resume screening without further fine-tuning.

Training Details

Training Data

The model was fine-tuned on a custom synthetic dataset consisting of over 150 rows. The data contrasts specific technical contributions (e.g., drone imagery processing, dataset curation, UI/UX design) against general engineering profiles and distinct extracurricular projects (e.g., BioMission, IdeavaultX).

Training Procedure

The model was trained using the Hugging Face Trainer API with the following hyperparameters:

  • Epochs: 3
  • Train Batch Size: 4
  • Evaluation Batch Size: 4
  • Optimizer: AdamW

Evaluation

Evaluation was conducted on a 20% holdout testing set. Due to the introduction of hard negatives (general students working on overlapping ML projects), the model avoids standard overfitting while maintaining high precision and recall for target identification.

Environmental Impact

  • Hardware Type: Google Colab Cloud instance (T4/V100 GPU)
  • Compute Region: Global
  • Framework: PyTorch & Transformers
Downloads last month
36
Safetensors
Model size
0.1B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support