PyTorch GNN Engineer with NLP Expertise for Paper Replication
We are seeking a skilled PyTorch GNN Engineer to assist in debugging and improving the replication of a research paper focused on Global and Local GCN for multi-label personality prediction. The ideal candidate will have experience with graph neural networks, natural language processing, and a solid understanding of model optimization techniques. You will work to identify issues, implement enhancements, and ensure the accuracy of results. this job is to replicating the paper “Knowledge-Enhanced Hierarchical Heterogeneous Graph for Personality”.
My code modules (already implemented)
• *Relevant Skills:**
- Strong PyTorch
- GNN/GCN experience, multi-label classification.
- Natural Language Processing (NLP)
- Model Debugging and Optimization
• *What I need **
Audit + Debug
-Confirm implementation matches paper (global Eq.1 + attention Eq.5 + contrastive loss).
-Find/fix issues (detach/no_grad, indexing, normalization, metric/threshold bugs).
-Improve training results
-Handle multi-label imbalance (pos_weight/focal, per-label thresholds on val).
-Stabilize training (LR, loss weighting BCE vs contrastive, clipping, seeds).
-Target: improve macro-F1 and avoid “F1=0” labels.
Apply Now
Apply Now
My code modules (already implemented)
• *Relevant Skills:**
- Strong PyTorch
- GNN/GCN experience, multi-label classification.
- Natural Language Processing (NLP)
- Model Debugging and Optimization
• *What I need **
Audit + Debug
-Confirm implementation matches paper (global Eq.1 + attention Eq.5 + contrastive loss).
-Find/fix issues (detach/no_grad, indexing, normalization, metric/threshold bugs).
-Improve training results
-Handle multi-label imbalance (pos_weight/focal, per-label thresholds on val).
-Stabilize training (LR, loss weighting BCE vs contrastive, clipping, seeds).
-Target: improve macro-F1 and avoid “F1=0” labels.
Apply Now
Apply Now