finding

active

finding:modified-cl-loss-outperforms-behavioral-das-loss-in-ood-transfer-from-dense-to-sparse-class-partition

Modified CL loss outperforms behavioral DAS loss in OOD transfer from dense to sparse class partition

Key practical utility result: CL loss improves generalization of alignment to out-of-distribution settings

Source paper

extracted_from

Addressing divergent representations from causal interventions on neural networks

(2025) · Satchel Grant · Simon Jerome Han · Alexa R. Tartaglini · Christopher Potts

Neighborhood — ranked by edge-count

Hypotheses (1)

hypothesis

We hypothesized that divergence could influence IIA when transferring the DAS alignment to OOD settings
supports
Motivating hypothesis for the OOD experiment testing practical utility of divergence reduction

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

DAS behavioral loss achieves IIA of 0.997±0.001 on synthetic 10-class dataset training/test setsfinding0.783
IIA baseline for DAS behavioral loss on synthetic dataset
Modified CL loss achieves IIA of 0.9988±0.0005 on synthetic 10-class dataset training/test setsfinding0.782
IIA for modified CL loss on synthetic dataset, comparable to behavioral DAS
For small CL loss weights epsilon, IIA is maintained (potentially improved) while EMD decreases in Boundless DAS on a 7B LLMfinding0.775
Empirical result showing the CL loss can reduce divergence without sacrificing interpretability accuracy
Modified CL loss produces EMD along feature dimensions of 0.007±0.001 on synthetic 10-class datasetfinding0.775
Quantitative improvement in divergence reduction using the modified CL loss on synthetic dataset
The modified CL loss is confined to a narrow set of simplistic settings and is not specific to pernicious divergenceconcept0.769
Explicitly identified limitation of the proposed mitigation method
DAS behavioral loss produces EMD along feature dimensions of 0.032±0.003 on synthetic 10-class datasetfinding0.761
Quantitative baseline for divergence using behavioral DAS loss on synthetic dataset
The CL auxiliary loss can directly reduce representational divergence in practical interpretability settings without sacrificing interpretability method accuracyclaim0.755
Central practical contribution: the CL loss offers a viable mitigation strategy
Modified CL Lossframework0.750
Novel variant of CL loss introduced in this paper targeting only causal subspace dimensions to improve OOD performance