Detecting out-of-distribution (OOD) inputs is a central challenge for safely deploying machine learning models in the real world. Existing solutions are mainly driven by small datasets, with low resolution and very few class labels (e.g., CIFAR). As a result, OOD detection for large-scale image classification tasks remains largely unexplored. In this paper, we bridge this critical gap by proposing a group-based OOD detection framework, along with a novel OOD scoring function termed MOS. Our key idea is to decompose the large semantic space into smaller groups with similar concepts, which allows simplifying the decision boundaries between in- vs. out-of-distribution data for effective OOD detection. Our method scales substantially better for high-dimensional class space than previous approaches. We evaluate models trained on ImageNet against four carefully curated OOD datasets, spanning diverse semantics. MOS establishes state-of-the-art performance, reducing the average FPR95 by 14.33% while achieving 6x speedup in inference compared to the previous best method.

MOS: Towards Scaling Out-of-distribution Detection for Large Semantic Space

Detecting out-of-distribution (OOD) inputs is critical for safely deploying deep learning models in an open-world setting. However, existing OOD detection solutions can be brittle in the open world, facing various types of adversarial OOD inputs. While methods leveraging auxiliary OOD data have emerged, our analysis on illuminative examples reveals a key insight that the majority of auxiliary OOD examples may not meaningfully improve or even hurt the decision boundary of the OOD detector, which is also observed in empirical results on real data. In this paper, we provide a theoretically motivated method, Adversarial Training with informative Outlier Mining (ATOM), which improves the robustness of OOD detection. We show that, by mining informative auxiliary OOD data, one can significantly improve OOD detection performance, and somewhat surprisingly, generalize to unseen adversarial attacks. ATOM achieves state-of-the-art performance under a broad family of classic and adversarial OOD evaluation tasks. For example, on the CIFAR-10 in-distribution dataset, ATOM reduces the FPR (at TPR 95%) by up to 57.99% under adversarial OOD inputs, surpassing the previous best baseline by a large margin.

ATOM: Robustifying Out-of-Distribution Detection Using Outlier Mining

Neural networks (NNs) are widely used for object classification in autonomous driving. However, NNs can fail on input data not well represented by the training dataset, known as out-of-distribution (OOD) data. A mechanism to detect OOD samples is important for safety-critical applications, such as automotive perception, to trigger a safe fallback mode. NNs often rely on softmax normalization for confidence estimation, which can lead to high confidences being assigned to OOD samples, thus hindering the detection of failures. This paper presents a method for determining whether inputs are OOD, which does not require OOD data during training and does not increase the computational cost of inference. The latter property is especially important in automotive applications with limited computational resources and real-time constraints. Our proposed approach outperforms state-of-the-art methods on real-world automotive datasets.

Out-of-Distribution Detection for Automotive Perception

/pdf/mos-towards-scaling-out-of-distribution-detection-for-large-249m0v00ui.pdf

Deep anomaly detection with self-supervised learning and adversarial training

Outlier exposure with confidence control for out-of-distribution detection

The technology of hyperspectral imaging (HSI) records the visual information upon long-range-distributed spectral wavelengths. A representative hyperspectral image acquisition procedure conducts a 3D-to-2D encoding by the coded aperture snapshot spectral imager (CASSI) and requires a software decoder for the 3D signal reconstruction. By observing this physical encoding procedure, two major challenges stand in the way of a high-fidelity reconstruction. (i) To obtain 2D measurements, CASSI dislocates multiple channels by disperser-titling and squeezes them onto the same spatial region, yielding an entangled data loss. (ii) The physical coded aperture leads to a masked data loss by selectively blocking the pixel-wise light exposure. To tackle these challenges, we propose a spatial-spectral (S^2-) Transformer network with a mask-aware learning strategy. First, we simultaneously leverage spatial and spectral attention modeling to disentangle the blended information in the 2D measurement along both two dimensions. A series of Transformer structures are systematically designed to fully investigate the spatial and spectral informative properties of the hyperspectral data. Second, the masked pixels will induce higher prediction difficulty and should be treated differently from unmasked ones. Thereby, we adaptively prioritize the loss penalty attributing to the mask structure by inferring the pixel-wise reconstruction difficulty upon the mask-encoded prediction. We theoretically discusses the distinct convergence tendencies between masked/unmasked regions of the proposed learning strategy. Extensive experiments demonstrates that the proposed method achieves superior reconstruction performance. Additionally, we empirically elaborate the behaviour of spatial and spectral attentions under the proposed architecture, and comprehensively examine the impact of the mask-aware learning.

S^2-Transformer for Mask-Aware Hyperspectral Image Reconstruction

Automated graph learning has drawn widespread research attention due to its great potential to reduce human efforts when dealing with graph data, among which hyperparameter optimization (HPO) is one of the mainstream directions and has made promising progress. However, how to obtain reliable and trustworthy prediction results with automated graph neural networks (GNN) is still quite underexplored. To this end, we investigate automated GNN calibration by marrying uncertainty estimation to the HPO problem. Specifically, we propose a hyperparameter uncertainty-induced graph convolutional network (HyperU-GCN) with a bilevel formulation, where the upper-level problem explicitly reasons uncertainties by developing a probabilistic hypernetworks through a variational Bayesian lens, while the lower-level problem learns how the GCN weights respond to a hyperparameter distribution. By squeezing model uncertainty into the hyperparameter space, the proposed HyperU-GCN could achieve calibrated predictions in a similar way to Bayesian model averaging over hyperparameters. Extensive experimental results on six public datasets were provided in terms of node classification accuracy and expected calibration error (ECE), demonstrating the effectiveness of our approach compared with several state-of-the-art uncertainty-aware and calibrated GCN methods.

Calibrate Automated Graph Neural Network via Hyperparameter Uncertainty

– Reconstruction Backbone SRN: A detailed introduction to the reconstruction backbone network used in the manuscript (Section 2). – Alternating Reconstruction Backbones: More analyses on alternating reconstruction backbones employed in the proposed method (Section 3). – Spectral Fidelity Analysis: Evaluation on the spectral fidelity of the reconstruction results by the proposed method (Section 4). – Epistemic Uncertainty Analysis: More visualization and discussion on epistemic uncertainty of the proposed method (Section 5). – Complementary Ablation Studies: Ablation studies under one-to-many miscalibration and the same mask setting (one-to-one) (Section 6). – Self-tuning variance Analysis: More discussions about the proposed selftuning variance. Specifically, more results for fixed variance are provided. Also, we demonstrate the convergence of gφ(m), variational noise distributions given distinct noise priors (Section 7). – Datasets: More illustrations on the dataset, includes training data, validation data, testing data, and mask set (Section 8). – Reconstruction Results: More reconstruction results under many-to-many setting (Section 9), one-to-many setting (Section 10) and the traditional setting (Section 11).

Modeling Mask Uncertainty in Hyperspectral Image Reconstruction (Supplementary Material)

The field of image super-resolution (SR) has witnessed extensive neural network designs from CNN to transformer architectures. However, prevailing SR models suffer from prohibitive memory footprint and intensive computations, which limits further deployment on computational-constrained platforms. In this work, we investigate the potential of network pruning for super-resolution to take advantage of off-the-shelf network designs and reduce the underlying computational overhead. Two main challenges remain in applying pruning methods for SR. First, the widely-used filter pruning technique reflects limited granularity and restricted adaptability to diverse network structures. Second, existing pruning methods generally operate upon a pre-trained network for the sparse structure determination, failing to get rid of dense model training in the traditional SR paradigm. To address these challenges, we adopt unstructured pruning with sparse models directly trained from scratch. Specifically, we propose a novel Iterative Soft Shrinkage-Percentage (ISS-P) method by optimizing the sparse structure of a randomly initialized network at each iteration and tweaking unimportant weights with a small amount proportional to the magnitude scale on-the-fly. We observe that the proposed ISS-P could dynamically learn sparse structures adapting to the optimization process and preserve the sparse model's trainability by yielding a more regularized gradient throughput. Experiments on benchmark datasets demonstrate the effectiveness of the proposed ISS-P compared with state-of-the-art methods over diverse network architectures.

Jiamian Wang

Papers

Outlier exposure with confidence control for out-of-distribution detection

S^2-Transformer for Mask-Aware Hyperspectral Image Reconstruction

Calibrate Automated Graph Neural Network via Hyperparameter Uncertainty

Modeling Mask Uncertainty in Hyperspectral Image Reconstruction (Supplementary Material)

Iterative Soft Shrinkage Learning for Efficient Image Super-Resolution