Embedded Files
Kervolution-Based SubSpectralNet for Acoustic Scene Classification
Kervolution-Based SubSpectralNet for Acoustic Scene Classification
Created a Kervolution-Based SubSpectralNet model for Acoustic Scene Classification. The proposed model increases non-linearity by introducing the kervolution operation and achieves an accuracy of 75.76% with the Gaussian kernel.
Multimodal Emotion Recognition Using Voice and Text
Multimodal Emotion Recognition Using Voice and Text
Developed a Deep Learning model on the MELD: Multimodal EmotionLines Dataset with TensorFlow that uses BERT and CNN to predict emotions using word embeddings and MFCCs to produce a joint output.
Word Embedding Models for Subjectivity Analysis
Word Embedding Models for Subjectivity Analysis
Subjectivity Analysis project on the Cornell SUBJ dataset, with the developed technique of BERT-based fine-tuning with Cyclic Learning Rate using TensorFlow being the current 2nd-best performing model on the task.
Geolocation Clustering for Resource Management in 6G Networks
Geolocation Clustering for Resource Management in 6G Networks
Created a hybrid HDBSCAN and KNN clustering methodology for 6G networks to enable optimal resource allocation by assigning the base stations to hot spots with heavy traffic requirements. Achieved a Silhouette Score of 0.92 and eliminated all outliers.
Page updated
Google Sites
Report abuse