Shaping Parameter Contribution Patterns for Out-of-Distribution Detection

Haonan Xu; Yang Yang

cs.LGMar 7, 2026

Shaping Parameter Contribution Patterns for Out-of-Distribution Detection

Authors: Haonan Xu, Yang Yang

Organizations: Nanjing University of Science and Technology

Abstract

Out-of-distribution (OOD) detection is a well-known challenge due to deep models often producing overconfident. In this paper, we reveal a key insight that trained classifiers tend to rely on sparse parameter contribution patterns, meaning that only a few dominant parameters drive predictions. This brittleness can be exploited by OOD inputs that anomalously trigger these parameters, resulting in overconfident predictions. To address this issue, we propose a simple yet effective method called Shaping Parameter Contribution Patterns (SPCP), which enhances OOD detection robustness by encouraging the classifier to learn boundary-oriented dense contribution patterns. Specifically, SPCP operates during training by rectifying excessively high parameter contributions based on a dynamically estimated threshold. This mechanism promotes the classifier to rely on a broader set of parameters for decision-making, thereby reducing the risk of overconfident predictions caused by anomalously triggered parameters, while preserving in-distribution (ID) performance. Extensive experiments under various OOD detection setups verify the effectiveness of SPCP.

Figures & tables

Panel of Figure 2 — Figure 2: An illustrative example of parameter contribution patterns in a ResNet-18 classifier trained on CIFAR-10 using cross-entropy loss. Units are sorted in the same order. This example focuses on the ‘airplane’ class, showcasing the contributions of each unit in the classifier weights to the model outputs for both ID and OOD samples. The parameter contributions are defined with reference to Eq. ( 4 ). Since model outputs are directly determined by parameter contributions, OOD input that anomalously triggers dominant parameters can be confidently yet incorrectly classified as ID categories and hurt OOD detection.

Panel of Figure 4 — Figure 4: Comparison of the classifier’s parameter contribution patterns before and after applying SPCP, with the average contribution matrix on the ID test set sorted for clarity.

Panel of Figure 6 — Figure 6: Comparison of normalized OOD score distributions before and after applying SPCP.

ID datasets		CIFAR-10		CIFAR-100
OOD datasets		FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$
		Vanilla training / SPCP training (Ours)
Near-OOD	CIFAR-10	–	–	59.21 $\pm$ 0.75 / 60.10 $\pm$ 0.95	79.05 $\pm$ 0.11 / 79.09 $\pm$ 0.22
	CIFAR-100	66.60 $\pm$ 4.46 / 35.74 $\pm$ 0.51	86.36 $\pm$ 0.58 / 90.42 $\pm$ 0.17	–	–
	TIN	56.08 $\pm$ 4.83 / 27.59 $\pm$ 0.48	88.80 $\pm$ 0.36 / 92.72 $\pm$ 0.07	52.03 $\pm$ 0.50 / 50.33 $\pm$ 0.85	82.76 $\pm$ 0.08 / 83.40 $\pm$ 0.28
	Average	61.34 $\pm$ 4.63 / 31.67 $\pm$ 0.25	87.58 $\pm$ 0.46 / 91.57 $\pm$ 0.11	55.62 $\pm$ 0.61 / 55.21 $\pm$ 0.85	80.91 $\pm$ 0.08 / 81.25 $\pm$ 0.19

Table 1: OOD detection performance on CIFAR benchmarks. All values are percentages, and the best results are in bold.

\uparrow

indicates that larger values are better,

\downarrow

indicates that smaller values are better.

ID datasets		ImageNet-200
OOD datasets		FPR95 $\downarrow$	AUROC $\uparrow$
		Vanilla training / SPCP training (Ours)
Near-OOD	SSB-hard	69.77 $\pm$ 0.32 / 70.23 $\pm$ 0.21	79.83 $\pm$ 0.02 / 79.80 $\pm$ 0.05
	NINCO	50.70 $\pm$ 0.89 / 49.32 $\pm$ 0.40	85.17 $\pm$ 0.11 / 85.24 $\pm$ 0.06
	Average	60.24 $\pm$ 0.57 / 59.77 $\pm$ 0.25	82.50 $\pm$ 0.05 / 82.52 $\pm$ 0.03
Far-OOD	iNaturalist	26.41 $\pm$ 2.29 / 23.86 $\pm$ 0.94	92.55 $\pm$ 0.50 / 92.91 $\pm$ 0.08

Table 2: OOD detection performance on large-scale ImageNet benchmark.

Method	CIFAR-10					CIFAR-100
	Near-OOD		Far-OOD		ID ACC $\uparrow$	Near-OOD		Far-OOD		ID ACC $\uparrow$
	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
	Post-hoc methods (vanilla training with cross-entropy)
MSP ( Hendrycks and Gimpel 2017 )	48.17 $\pm$ 3.92	88.03 $\pm$ 0.25	31.72 $\pm$ 1.84	90.73 $\pm$ 0.43	95.06 $\pm$ 0.30	54.80 $\pm$ 0.33	80.27 $\pm$ 0.11	58.70 $\pm$ 1.06	77.76 $\pm$ 0.44	77.25 $\pm$ 0.10
Energy ( Liu et al. 2020 )	61.34 $\pm$ 4.63	87.58 $\pm$ 0.46	41.69 $\pm$ 5.32	91.21 $\pm$ 0.92	95.06 $\pm$ 0.30	55.62 $\pm$ 0.61	80.91 $\pm$ 0.08	56.59 $\pm$ 1.38	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
DICE ( Sun and Li 2022 )	70.04 $\pm$ 7.64	78.34 $\pm$ 0.79	51.76 $\pm$ 4.42	84.23 $\pm$ 1.89	95.06 $\pm$ 0.30	57.95 $\pm$ 0.53	79.38 $\pm$ 0.23	56.25 $\pm$ 0.60	80.01 $\pm$ 0.18	77.25 $\pm$ 0.10

Table 3: Comparison on CIFAR benchmarks. All values are percentages, and OOD detection results are averaged over multiple OOD datasets. Detailed results for each OOD dataset are provided in Appendix D. The best results are in bold , with the second and third best results underlined .

Method	Near-OOD		Far-OOD		ID ACC $\uparrow$
Method	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
	Post-hoc methods (vanilla training with cross-entropy)
MSP ( Hendrycks and Gimpel 2017 )	54.82 $\pm$ 0.35	83.34 $\pm$ 0.06	35.43 $\pm$ 0.38	90.13 $\pm$ 0.09	86.37 $\pm$ 0.08
Energy ( Liu et al. 2020 )	60.24 $\pm$ 0.57	82.50 $\pm$ 0.05	34.86 $\pm$ 1.30	90.86 $\pm$ 0.21	86.37 $\pm$ 0.08
DICE ( Sun and Li 2022 )	61.88 $\pm$ 0.67	81.78 $\pm$ 0.14	36.51 $\pm$ 1.18	90.80 $\pm$ 0.31	86.37 $\pm$ 0.08
ReAct ( Sun et al. 2021 )	62.49 $\pm$ 2.19	81.87 $\pm$ 0.98	28.50 $\pm$ 0.95	92.31 $\pm$ 0.56	86.37 $\pm$ 0.08

Table 4: Comparison on large-scale ImageNet benchmark.

Stage		Near-OOD		Far-OOD		ID ACC $\uparrow$
Training	Inferring	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
$\times$	$\times$	55.62 $\pm$ 0.61	80.91 $\pm$ 0.08	56.59 $\pm$ 1.38	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
$\checkmark$	$\times$	55.49 $\pm$ 0.77	81.09 $\pm$ 0.21	51.96 $\pm$ 2.25	81.71 $\pm$ 1.45	77.70 $\pm$ 0.24
$\times$	$\checkmark$	60.28 $\pm$ 0.46	79.50 $\pm$ 0.08	53.66 $\pm$ 3.13	80.59 $\pm$ 1.10	76.94 $\pm$ 0.12
$\checkmark$	$\checkmark$	55.21 $\pm$ 0.85	81.25 $\pm$ 0.19	50.59 $\pm$ 1.83	82.17 $\pm$ 1.32	77.70 $\pm$ 0.20

Table 5: Ablation study of contribution truncation in different stages on the CIFAR-100 benchmark.

Panel of Figure 7 — Figure 7: Ablation studies on the hyperparameters: (a) effect of varying the percentile $\rho$ used for threshold estimation; (b) impact of the EMA smoothing factor $\beta$ ; (c) influence of initial threshold $\lambda_{0}$ . The OOD results are averaged over both near- and far-OOD groups on the CIFAR-100 benchmark.

Model	Method	Near-OOD		Far-OOD		ID ACC $\uparrow$
Model	Method	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
ResNet-18	Vanilla	55.62 $\pm$ 0.61	80.91 $\pm$ 0.08	56.59 $\pm$ 1.38	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
ResNet-18	SPCP	55.21 $\pm$ 0.85	81.25 $\pm$ 0.19	50.59 $\pm$ 1.83	82.17 $\pm$ 1.32	77.70 $\pm$ 0.20
WideResNet-28-10	Vanilla	54.48 $\pm$ 1.30	82.31 $\pm$ 0.41	55.34 $\pm$ 1.67	80.87 $\pm$ 0.80	80.53 $\pm$ 0.13
WideResNet-28-10	SPCP	53.48 $\pm$ 0.52	82.63 $\pm$ 0.24	52.84 $\pm$ 0.93	81.69 $\pm$ 0.31	80.56 $\pm$ 0.22
DenseNet-101	Vanilla	61.57 $\pm$ 1.36	79.27 $\pm$ 0.36	65.49 $\pm$ 2.45	76.16 $\pm$ 0.81	76.54 $\pm$ 0.42

Table 6: Generalization to different backbones on the CIFAR-100 benchmark.

Method	Near-OOD		Far-OOD		ID ACC $\uparrow$
Method	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
MSP ( Hendrycks and Gimpel 2017 )	54.80 $\pm$ 0.33	80.27 $\pm$ 0.11	58.70 $\pm$ 1.06	77.76 $\pm$ 0.44	77.25 $\pm$ 0.10
MSP+SPCP	55.04 $\pm$ 0.68	80.41 $\pm$ 0.28	54.31 $\pm$ 0.95	80.22 $\pm$ 0.93	77.70 $\pm$ 0.20
Energy ( Liu et al. 2020 )	55.62 $\pm$ 0.61	80.91 $\pm$ 0.08	56.59 $\pm$ 1.38	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
Energy+SPCP	55.21 $\pm$ 0.85	81.25 $\pm$ 0.19	50.59 $\pm$ 1.83	82.17 $\pm$ 1.32	77.70 $\pm$ 0.20
GradNorm ( Huang et al. 2021 )	85.58 $\pm$ 0.46	70.13 $\pm$ 0.47	83.68 $\pm$ 1.92	69.14 $\pm$ 1.05	77.25 $\pm$ 0.10

Table 7: Compatibility with other OOD detection methods on the CIFAR-100 benchmark.

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Panel of Figure 9 — Figure 9: Visualization of the classifier’s parameter contribution patterns, with the average contribution matrix on the ID test set sorted for improved clarity.

Method	Near-OOD		Far-OOD		ID ACC $\uparrow$
Method	FPR95 $\downarrow$	AUROC $\uparrow$	FPR95 $\downarrow$	AUROC $\uparrow$	ID ACC $\uparrow$
Vanilla	55.62 $\pm$ 0.61	80.91 $\pm$ 0.08	56.59 $\pm$ 1.38	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
+ $L_{2}$ (0.1)	63.48 $\pm$ 0.85	79.13 $\pm$ 0.37	65.03 $\pm$ 2.27	78.19 $\pm$ 0.59	76.90 $\pm$ 0.55
+ $L_{2}$ (0.01)	56.66 $\pm$ 1.05	80.75 $\pm$ 0.15	57.20 $\pm$ 1.28	79.34 $\pm$ 1.17	78.31 $\pm$ 0.21
+ $L_{2}$ (0.001)	56.54 $\pm$ 1.35	80.65 $\pm$ 0.32	53.93 $\pm$ 1.93	80.70 $\pm$ 1.18	77.60 $\pm$ 0.31
+ $L_{2}$ (0.0001)	55.87 $\pm$ 1.32	80.93 $\pm$ 0.38	54.00 $\pm$ 1.84	80.85 $\pm$ 0.97	77.40 $\pm$ 0.28

Appendix

Table 8: Comparison of

L_{2}

weight regularization and SPCP on the CIFAR-100 benchmark.

Dataset	Methods	Latency (s)
Dataset	Methods	Training	Inferring
CIFAR-10	Vanilla	7.82	2.35
	+LogitNorm	7.99	2.37
	+SPCP	8.19	2.48
ImageNet-200	Vanilla	515.71	10.94
	+LogitNorm	518.43	10.97

Appendix

Table 9: Analysis of the training and inference overhead introduced by SPCP. This overhead is measured by latency (

\downarrow

), defined as the time required to process the entire training set during training and the entire test set during inference in a single run. Lower latency indicates reduced computational cost. All experiments are conducted under consistent software and hardware settings as detailed in Appendix C , with results averaged over multiple independent runs.

	CIFAR-10		ImageNet-200
	Near-OOD	Far-OOD	Near-OOD	Far-OOD
ITP	52.72 / 87.31	35.70 / 90.30	60.09 / 82.68	32.17 / 91.28
SPCP	31.67 / 91.57	20.54 / 94.55	59.77 / 82.52	30.43 / 91.46

Appendix

Table 10: Comparison of OOD Detection Performance between SPCP and ITP.

	Near-OOD	Far-OOD
Fixed	35.19 / 90.86	26.69 / 92.53
Ours	31.67 / 91.57	20.54 / 94.55

Appendix

Table 11: Ablation study on the adaptive

\lambda

conducted on the CIFAR-10 benchmark. Results (FPR95

\downarrow

/ AUROC

\uparrow

) are averaged over multiple runs.

	Kinetics-Sound
Vanilla	76.95 / 69.67
Ours	69.20 / 73.48

Appendix

Table 12: Extend SPCP to audio OOD detection tasks. Results (FPR95

\downarrow

/ AUROC

\uparrow

) are averaged over multiple runs.

\lambda=\beta\cdot\lambda+(1-\beta)\cdot\frac{1}{|\mathcal{B}|}\sum_{\mathbf{x}_{i}\in\mathcal{B}}\text{Top}(\rho,C(\mathbf{x}_{i}))

Appendix

Algorithm 1 Training Procedure of SPCP

Method	Near-OOD			Far-OOD					ID ACC
Method	CIFAR-100	TIN	Average	MNIST	SVHN	Textures	Places365	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	87.19 $\pm$ 0.33	88.87 $\pm$ 0.19	88.03 $\pm$ 0.25	92.63 $\pm$ 1.57	91.46 $\pm$ 0.40	89.89 $\pm$ 0.71	88.92 $\pm$ 0.47	90.73 $\pm$ 0.43	95.06 $\pm$ 0.30
Energy	86.36 $\pm$ 0.58	88.80 $\pm$ 0.36	87.58 $\pm$ 0.46	94.32 $\pm$ 2.53	91.79 $\pm$ 0.98	89.47 $\pm$ 0.70	89.25 $\pm$ 0.78	91.21 $\pm$ 0.92	95.06 $\pm$ 0.30
DICE	77.01 $\pm$ 0.88	79.67 $\pm$ 0.87	78.34 $\pm$ 0.79	90.37 $\pm$ 5.97	90.02 $\pm$ 1.77	81.86 $\pm$ 2.35	74.67 $\pm$ 4.98	84.23 $\pm$ 1.89	95.06 $\pm$ 0.30
ReAct	85.93 $\pm$ 0.83	88.29 $\pm$ 0.44	87.11 $\pm$ 0.61	92.81 $\pm$ 3.03	89.12 $\pm$ 3.19	89.38 $\pm$ 1.49	90.35 $\pm$ 0.78	90.42 $\pm$ 1.41	95.06 $\pm$ 0.30

Appendix

Table 13: Fine-grained results (AUROC

\uparrow

) on the CIFAR-10 benchmark.

Method	Near-OOD			Far-OOD					ID ACC
Method	CIFAR-100	TIN	Average	MNIST	SVHN	Textures	Places365	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	53.08 $\pm$ 4.86	43.27 $\pm$ 3.00	48.17 $\pm$ 3.92	23.64 $\pm$ 5.81	25.82 $\pm$ 1.64	34.96 $\pm$ 4.64	42.47 $\pm$ 3.81	31.72 $\pm$ 1.84	95.06 $\pm$ 0.30
Energy	66.60 $\pm$ 4.46	56.08 $\pm$ 4.83	61.34 $\pm$ 4.63	24.99 $\pm$ 12.93	35.12 $\pm$ 6.11	51.82 $\pm$ 6.11	54.85 $\pm$ 6.52	41.69 $\pm$ 5.32	95.06 $\pm$ 0.30
DICE	73.71 $\pm$ 7.67	66.37 $\pm$ 7.68	70.04 $\pm$ 7.64	30.83 $\pm$ 10.54	36.61 $\pm$ 4.74	62.42 $\pm$ 4.79	77.19 $\pm$ 12.60	51.76 $\pm$ 4.42	95.06 $\pm$ 0.30
ReAct	67.40 $\pm$ 7.34	59.71 $\pm$ 7.31	63.56 $\pm$ 7.33	33.77 $\pm$ 18.00	50.23 $\pm$ 15.98	51.42 $\pm$ 11.42	44.20 $\pm$ 3.35	44.90 $\pm$ 8.37	95.06 $\pm$ 0.30

Appendix

Table 14: Fine-grained results (FPR95

\downarrow

) on the CIFAR-10 benchmark.

Method	Near-OOD			Far-OOD					ID ACC
Method	CIFAR-10	TIN	Average	MNIST	SVHN	Textures	Places365	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	78.47 $\pm$ 0.07	82.07 $\pm$ 0.17	80.27 $\pm$ 0.11	76.08 $\pm$ 1.86	78.42 $\pm$ 0.89	77.32 $\pm$ 0.71	79.22 $\pm$ 0.29	77.76 $\pm$ 0.44	77.25 $\pm$ 0.10
Energy	79.05 $\pm$ 0.11	82.76 $\pm$ 0.08	80.91 $\pm$ 0.08	79.18 $\pm$ 1.37	82.03 $\pm$ 1.74	78.35 $\pm$ 0.83	79.52 $\pm$ 0.23	79.77 $\pm$ 0.61	77.25 $\pm$ 0.10
DICE	78.04 $\pm$ 0.32	80.72 $\pm$ 0.30	79.38 $\pm$ 0.23	79.86 $\pm$ 1.89	84.22 $\pm$ 2.00	77.63 $\pm$ 0.34	78.33 $\pm$ 0.66	80.01 $\pm$ 0.18	77.25 $\pm$ 0.10
ReAct	78.65 $\pm$ 0.05	82.88 $\pm$ 0.08	80.77 $\pm$ 0.05	78.37 $\pm$ 1.59	83.01 $\pm$ 0.97	80.15 $\pm$ 0.46	80.03 $\pm$ 0.11	80.39 $\pm$ 0.49	77.25 $\pm$ 0.10

Appendix

Table 15: Fine-grained results (AUROC

\uparrow

) on the CIFAR-100 benchmark.

Method	Near-OOD			Far-OOD					ID ACC
Method	CIFAR-10	TIN	Average	MNIST	SVHN	Textures	Places365	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	58.91 $\pm$ 0.93	50.70 $\pm$ 0.34	54.80 $\pm$ 0.33	57.23 $\pm$ 4.68	59.07 $\pm$ 2.53	61.88 $\pm$ 1.28	56.62 $\pm$ 0.87	58.70 $\pm$ 1.06	77.25 $\pm$ 0.10
Energy	59.21 $\pm$ 0.75	52.03 $\pm$ 0.50	55.62 $\pm$ 0.61	52.62 $\pm$ 3.83	53.62 $\pm$ 3.14	62.35 $\pm$ 2.06	57.75 $\pm$ 0.86	56.59 $\pm$ 1.38	77.25 $\pm$ 0.10
DICE	60.98 $\pm$ 1.10	54.93 $\pm$ 0.53	57.95 $\pm$ 0.53	51.79 $\pm$ 3.67	49.58 $\pm$ 3.32	64.23 $\pm$ 1.65	59.39 $\pm$ 1.25	56.25 $\pm$ 0.60	77.25 $\pm$ 0.10
ReAct	61.30 $\pm$ 0.43	51.47 $\pm$ 0.47	56.39 $\pm$ 0.34	56.04 $\pm$ 5.66	50.41 $\pm$ 2.02	55.04 $\pm$ 0.82	55.30 $\pm$ 0.41	54.20 $\pm$ 1.56	77.25 $\pm$ 0.10

Appendix

Table 16: Fine-grained results (FPR95

\downarrow

) on the CIFAR-100 benchmark.

Method	Near-OOD			Far-OOD				ID ACC
Method	SSB-hard	NINCO	Average	iNaturalist	Textures	OpenImage-O	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	80.38 $\pm$ 0.03	86.29 $\pm$ 0.11	83.34 $\pm$ 0.06	92.80 $\pm$ 0.25	88.36 $\pm$ 0.13	89.24 $\pm$ 0.02	90.13 $\pm$ 0.09	86.37 $\pm$ 0.08
Energy	79.83 $\pm$ 0.02	85.17 $\pm$ 0.11	82.50 $\pm$ 0.05	92.55 $\pm$ 0.50	90.79 $\pm$ 0.16	89.23 $\pm$ 0.26	90.86 $\pm$ 0.21	86.37 $\pm$ 0.08
DICE	79.06 $\pm$ 0.05	84.49 $\pm$ 0.24	81.78 $\pm$ 0.14	91.81 $\pm$ 0.79	91.53 $\pm$ 0.21	89.06 $\pm$ 0.34	90.80 $\pm$ 0.31	86.37 $\pm$ 0.08
ReAct	78.97 $\pm$ 1.33	84.76 $\pm$ 0.64	81.87 $\pm$ 0.98	93.65 $\pm$ 0.88	92.86 $\pm$ 0.47	90.40 $\pm$ 0.35	92.31 $\pm$ 0.56	86.37 $\pm$ 0.08

Appendix

Table 17: Fine-grained results (AUROC

\uparrow

) on the ImageNet-200 benchmark.

Method	Near-OOD			Far-OOD				ID ACC
Method	SSB-hard	NINCO	Average	iNaturalist	Textures	OpenImage-O	Average	ID ACC
Post-hoc methods (vanilla training with cross-entropy)
MSP	66.00 $\pm$ 0.10	43.65 $\pm$ 0.75	54.82 $\pm$ 0.35	26.48 $\pm$ 0.73	44.58 $\pm$ 0.68	35.23 $\pm$ 0.18	35.43 $\pm$ 0.38	86.37 $\pm$ 0.08
Energy	69.77 $\pm$ 0.32	50.70 $\pm$ 0.89	60.24 $\pm$ 0.57	26.41 $\pm$ 2.29	41.43 $\pm$ 1.85	36.74 $\pm$ 1.14	34.86 $\pm$ 1.30	86.37 $\pm$ 0.08
DICE	70.84 $\pm$ 0.30	52.91 $\pm$ 1.20	61.88 $\pm$ 0.67	29.66 $\pm$ 2.62	40.96 $\pm$ 1.87	38.91 $\pm$ 1.16	36.51 $\pm$ 1.18	86.37 $\pm$ 0.08
ReAct	71.51 $\pm$ 1.92	53.47 $\pm$ 2.46	62.49 $\pm$ 2.19	22.97 $\pm$ 2.25	29.67 $\pm$ 1.35	32.86 $\pm$ 0.74	28.50 $\pm$ 0.95	86.37 $\pm$ 0.08

Appendix

Table 18: Fine-grained results (FPR95

\downarrow

) on the ImageNet-200 benchmark.

OOD Detection

Out-of-distribution (OOD) detection identifies inputs unlike a model’s training data so systems can flag uncertainty, reject inputs, or adapt safely.

See all 223 papers

Sparse Autoencoders for Interpretable Out-of-Distribution Detection
Jul 13, 2026Ayush Karmacharya, Luke Luschwitz, Lucia Romero +2OOD Detection
Purdue University · EPFL
Optimizing Three Critical Factors for Practical and Effective OOD Detection Fine-Tuning
Aug 2, 2023Hyunjun Choi, JaeHo Chung, Hawook JeongFine-Tuning OOD Detection
ASRI, ECE., Seoul National University, South Korea · RideFlux Inc., South Korea
AOE: Exhaustive Out-of-Distribution Detection via Recalibrating Outlier Labels
May 27, 2026Fengqiang Wan, Qing-Yuan Jiang, Yang YangOOD Detection
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, China.

Jul 13, 2026J/K move · Enter open · S save

Sparse Autoencoders for Interpretable Out-of-Distribution Detection

Ayush Karmacharya, Luke Luschwitz, Lucia Romero, Yanan Niu+1

Purdue University · EPFL

OOD Detection

Open paper AskPDF arXiv

Reliable detection of out-of-distribution (OOD) samples is crucial for the safe deployment of machine learning models. Neural networks often produce overconfident predictions for inputs that deviate from their training data, leading to significant degradation in performance. While many OOD detection methods focus on the final output layer, they neglect the rich hierarchical information present in intermediate network layers. This paper introduces a novel approach that leverages sparse autoencoders (SAEs) to learn interpretable features from these intermediate activations. We find that in-distribution (ID) and OOD data activate distinct sets of these sparse features. We propose a new OOD score derived from the cosine similarity between the sparse feature activations of a test sample and the mean activations of ID classes. Our post-hoc detection method not only achieves state-of-the-art performance on standard OOD detection benchmarks, but yields interpretable insights into how distribution shift affects learned representations.

Abstract

OOD Detection

Explore similar work

Sparse Autoencoders for Interpretable Out-of-Distribution Detection

Optimizing Three Critical Factors for Practical and Effective OOD Detection Fine-Tuning

AOE: Exhaustive Out-of-Distribution Detection via Recalibrating Outlier Labels