Stratis Ioannidis, Jennifer Dy, Lukas Miklautz, Andrii Shkabrii, Claudia Plant, Chengzhi Shi, Theodoros Thirimachos Davarakis, Prudence Lam
We lifted 20 functions out of this paper's own repositories and ran 16 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| neu-spiral/H-SPLID | — | 14 of 18 |
| copy not recorded | — | 2 of 2 |
| Function | Status | Where it lives |
|---|---|---|
| GradualWarmupScheduler | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("db476832a8e53425") |
| StandardClassificationModel | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("90a392a6f1cce5e6") |
| calculate_optimal_beta_weights_special_case | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("986760073a8f5ef4") |
| compression_loss | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("c1d639c433616c1b") |
| compute_centers | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("f3bccda2e14d7253") |
| distmat | Ran | this paper's copy was not recorded; identical code first harvested from sjyucnel/cauchy-schwarz-information-bottleneck pointer only · get_code("ca565da6831e8ade") |
| get_hard_assignments | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("5960316c22a26fdd") |
| hsic_normalized_cca | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("f9f266ad1f605127") |
| hsic_split_objective | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("104983bddcc567f7") |
| int_to_one_hot | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("22cae19b9320887c") |
| kernelmat | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("b41642f800673cde") |
| optimal_beta | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("fb86b08cc83c2b7a") |
| set_optimizer | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("7d331d112a46a892") |
| sigma_estimation | Ran | this paper's copy was not recorded; identical code first harvested from sjyucnel/cauchy-schwarz-information-bottleneck pointer only · get_code("20b89645c8dfbf6d") |
| split_embeddings | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("df1efb712bb815a9") |
| squared_euclidean_distance | Ran | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("cd32c1472fdae85b") |
| SplitTrain | Not yet run | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("0fe94fb1b541e6aa") |
| acedec_loss | Not yet run | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("06dffd9dd5c45ae3") |
| get_beta_weights | Not yet run | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("baf1fe01724958c5") |
| update_centers_ | Not yet run | neu-spiral/H-SPLID/src/models/split_model.py code served (permissive licence) · get_code("093a1c364685e7f1") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
We introduce H-SPLID, a novel algorithm for learning salient feature representations through the explicit decomposition of salient and non-salient features into separate spaces. We show that H-SPLID promotes learning low-dimensional, task-relevant features. We prove that the expected prediction deviation under input perturbations is upper-bounded by the dimension of the salient subspace and the Hilbert-Schmidt Independence Criterion (HSIC) between inputs and representations. This establishes a link between robustness and latent representation compression in terms of the dimensionality and information preserved. Empirical evaluations on image classification tasks show that models trained with H-SPLID primarily rely on salient input components, as indicated by reduced sensitivity to perturbations affecting non-salient features, such as image backgrounds. * Equal contribution. ‡ Shared supervision. † Main work done during a research stay at Northeastern University. 39th Conference on Neural Information Processing Systems (NeurIPS 2025).
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2510.20627")
get_code_for_paper("2510.20627")
have("2510.20627")
Connect an agent — have() is free.