| SparCity innovation | Description | SparCity WP | Benefiting usecase(s) |
| Novel usage of Graphcore IPUs | New techniques of programming, profiling, optimization for IPUs | WP2 | Usecase 1, Usecase 2, Usecase 3, Usecase 4, Usecase 6 |
| Non-standard partitioning/balancing strategies | Multi-objective and/or hierarchical graph partitioning | WP2 | Usecase 1, Usecase 5, Usecase 6 |
| ML-supported recommendation of re-ordering schemes | Use of pattern characteristics of sparse matrices for guidance in choosing re-ordering schemes | WP1 | Usecase 1, Usecase 3 |
| A software library for automated offloading of MPI communication | Use of dedicated OpenMP threads for handling communication tasks | WP3 | Usecase 1 |
| Data and computation integration | Fuse of GPU kernels with similar data accesses to save memory traffic | WP2 | Usecase 2 |
| Sparse computation aware performance modeling | Incorporation of irregular memory traffic in performance modeling | WP1 | Usecase 1, Usecase 3 |
| Performance dashboard using SuperTwin | Real-time performance monitoring and analysis | WP4 | Usecase 1, Usecase 3 |
Usecase 2: Detection of digital wildfires in social media
Usecase 3: High-order epistasis detection
Usecase 4: Multiple pedestrian tracking in autonomous driving
Usecase 5: Sparsification of deep-learning models
Usecase 6: Sequence alignment