Mitotic figure detection in histology images is a hard-to-define, yet clinically significant task, where labels are generated with pathologist interpretations and where there is no ``gold-standard'' independent ground-truth. However, it is well-established that these interpretation based labels are often unreliable, in part, due to differences in expertise levels and human subjectivity. In this paper, our goal is to shed light on the inherent uncertainty of mitosis labels and characterize the mitotic figure classification task in a human interpretable manner. We train a probabilistic diffusion model to synthesize patches of cell nuclei for a given mitosis label condition. Using this model, we can then generate a sequence of synthetic images that correspond to the same nucleus transitioning into the mitotic state. This allows us to identify different image features associated with mitosis, such as cytoplasm granularity, nuclear density, nuclear irregularity and high contrast between the nucleus and the cell body. Our approach offers a new tool for pathologists to interpret and communicate the features driving the decision to recognize a mitotic figure.
One potential drawback of using aggregated performance measurement in machine learning is that models may learn to accept higher errors on some training cases as compromises for lower errors on others, with the lower errors actually being instances of overfitting. This can lead to both stagnation at local optima and poor generalization. Lexicase selection is an uncompromising method developed in evolutionary computation, which selects models on the basis of sequences of individual training case errors instead of using aggregated metrics such as loss and accuracy. In this paper, we investigate how lexicase selection, in its general form, can be integrated into the context of deep learning to enhance generalization. We propose Gradient Lexicase Selection, an optimization framework that combines gradient descent and lexicase selection in an evolutionary fashion. Our experimental results demonstrate that the proposed method improves the generalization performance of various widely-used deep neural network architectures across three image classification benchmarks. Additionally, qualitative analysis suggests that our method assists networks in learning more diverse representations. Our source code is available on GitHub: https://github.com/ld-ing/gradient-lexicase.
Channel attention mechanisms endeavor to recalibrate channel weights to enhance representation abilities of networks. However, mainstream methods often rely solely on global average pooling as the feature squeezer, which significantly limits the overall potential of models. In this paper, we investigate the statistical moments of feature maps within a neural network. Our findings highlight the critical role of high-order moments in enhancing model capacity. Consequently, we introduce a flexible and comprehensive mechanism termed Extensive Moment Aggregation (EMA) to capture the global spatial context. Building upon this mechanism, we propose the Moment Channel Attention (MCA) framework, which efficiently incorporates multiple levels of moment-based information while minimizing additional computation costs through our Cross Moment Convolution (CMC) module. The CMC module via channel-wise convolution layer to capture multiple order moment information as well as cross channel features. The MCA block is designed to be lightweight and easily integrated into a variety of neural network architectures. Experimental results on classical image classification, object detection, and instance segmentation tasks demonstrate that our proposed method achieves state-of-the-art results, outperforming existing channel attention methods.
Downsampling layers are crucial building blocks in CNN architectures, which help to increase the receptive field for learning high-level features and reduce the amount of memory/computation in the model. In this work, we study the generalization of the uniform downsampling layer for group equivariant architectures, e.g., G-CNNs. That is, we aim to downsample signals (feature maps) on general finite groups with anti-aliasing. This involves the following: (a) Given a finite group and a downsampling rate, we present an algorithm to form a suitable choice of subgroup. (b) Given a group and a subgroup, we study the notion of bandlimited-ness and propose how to perform anti-aliasing. Notably, our method generalizes the notion of downsampling based on classical sampling theory. When the signal is on a cyclic group, i.e., periodic, our method recovers the standard downsampling of an ideal low-pass filter followed by a subsampling operation. Finally, we conducted experiments on image classification tasks demonstrating that the proposed downsampling operation improves accuracy, better preserves equivariance, and reduces model size when incorporated into G-equivariant networks
Machine learning has achieved remarkable advancements but at the cost of significant computational resources. This has created an urgent need for a novel and energy-efficient computational fabric. CMOS Oscillator Networks (OscNet) is a brain inspired and specially designed hardware for low energy consumption. In this paper, we propose a Hopfield Network based machine learning algorithm that can be implemented on OscNet. The network is trained using forward propagation alone to learn sparsely connected weights, yet achieves an 8% improvement in accuracy compared to conventional deep learning models on MNIST dataset. OscNet v1.5 achieves competitive accuracy on MNIST and is well-suited for implementation using CMOS-compatible ring oscillator arrays with SHIL. In oscillator-based implementation, we utilize only 24% of the connections used in a fully connected Hopfield network, with merely a 0.1% drop in accuracy. OscNet v1.5 relies solely on forward propagation and employs sparse connections, making it an energy-efficient machine learning pipeline designed for CMOS oscillator computing. The repository for OscNet family is: https://github.com/RussRobin/OscNet.
약지도 객체 위치 추정(Weakly Supervised Object Localization, WSOL)은 전역 클래스 수준 레이블만을 사용하여 분류 및 위치 추정(LOC)을 위한 심층 학습 모델을 학습할 수 있게 한다. 학습 중 경계 상자(bounding box, bbox) 지도 정보의 부재는 하이퍼파라미터 튜닝, 모델 선택, 평가 측면에서 문헌상 여러 과제를 야기한다. WSOL 방법들은 모델 선택을 위해 경계 상자 어노테이션이 포함된 검증 세트에, 그리고 위치 추정 맵으로부터 경계 상자를 생성하기 위한 임계값 추정에 경계 상자 어노테이션이 포함된 테스트 세트에 의존한다. 그러나 이 접근 방식은 실제 환경에서 해당 어노테이션을 일반적으로 구할 수 없다는 점에서 WSOL 설정과 부합하지 않는다. 초기 실증 분석에서는 모델 선택과 임계값 추정을 각각 클래스 레이블과 이미지 자체에만 의존할 경우, 수동 경계 상자 어노테이션을 사용하는 경우에 비해 위치 추정 성능이 크게 저하됨을 확인하였다. 이는 최적의 모델 성능을 위해 경계 상자 레이블을 활용하는 것의 중요성을 강조한다. 본 논문에서는 수동 경계 상자 어노테이션 없이 위치 추정 정보를 제공하는 새로운 WSOL 평가 프로토콜을 제안한다. 구체적으로, Selective Search, CLIP, RPN과 같은 사전학습된 기성 영역 제안(region proposal) 방법으로부터 노이즈가 포함된 의사 경계 상자(pseudo-box)를 생성하여 모델 선택에 활용하였다. 이러한 경계 상자는 위치 추정 맵에서 임계값을 추정하는 데에도 사용되어 테스트 세트 경계 상자 어노테이션의 필요성을 우회한다. 도전적인 자연 이미지 및 의료 이미지 데이터셋에서 여러 WSOL 방법으로 수행한 실험 결과, 제안하는 의사 경계 상자를 검증에 활용하면 모델 선택과 임계값 추정이 용이해지며, 검증 세트에서 GT 경계 상자로 선택하고 테스트 세트에서 임계값을 추정한 모델과 비슷한 위치 추정 성능을 달성함을 보였다. 또한 클래스 수준 레이블로 선택한 후 위치 추정 맵만으로 동적 임계값을 적용한 모델보다 우수한 성능을 나타냈다.
Slimmable networks are a family of neural networks that can instantly adjust the runtime width. The width can be chosen from a predefined widths set to adaptively optimize accuracy-efficiency trade-offs at runtime. In this work, we propose a systematic approach to train universally slimmable networks (US-Nets), extending slimmable networks to execute at arbitrary width, and generalizing to networks both with and without batch normalization layers. We further propose two improved training techniques for US-Nets, named the sandwich rule and inplace distillation, to enhance training process and boost testing accuracy. We show improved performance of universally slimmable MobileNet v1 and MobileNet v2 on ImageNet classification task, compared with individually trained ones and 4-switch slimmable network baselines. We also evaluate the proposed US-Nets and improved training techniques on tasks of image super-resolution and deep reinforcement learning. Extensive ablation experiments on these representative tasks demonstrate the effectiveness of our proposed methods. Our discovery opens up the possibility to directly evaluate FLOPs-Accuracy spectrum of network architectures. Code and models are available at: https://github.com/JiahuiYu/slimmable_networks
In order to efficiently learn with small amount of data on new tasks, meta-learning transfers knowledge learned from previous tasks to the new ones. However, a critical challenge in meta-learning is the task heterogeneity which cannot be well handled by traditional globally shared meta-learning methods. In addition, current task-specific meta-learning methods may either suffer from hand-crafted structure design or lack the capability to capture complex relations between tasks. In this paper, motivated by the way of knowledge organization in knowledge bases, we propose an automated relational meta-learning (ARML) framework that automatically extracts the cross-task relations and constructs the meta-knowledge graph. When a new task arrives, it can quickly find the most relevant structure and tailor the learned structure knowledge to the meta-learner. As a result, the proposed framework not only addresses the challenge of task heterogeneity by a learned meta-knowledge graph, but also increases the model interpretability. We conduct extensive experiments on 2D toy regression and few-shot image classification and the results demonstrate the superiority of ARML over state-of-the-art baselines.
Learning rate schedulers have been widely adopted in training deep neural networks. Despite their practical importance, there is a discrepancy between its practice and its theoretical analysis. For instance, it is not known what schedules of SGD achieve best convergence, even for simple problems such as optimizing quadratic objectives. In this paper, we propose Eigencurve, the first family of learning rate schedules that can achieve minimax optimal convergence rates (up to a constant) for SGD on quadratic objectives when the eigenvalue distribution of the underlying Hessian matrix is skewed. The condition is quite common in practice. Experimental results show that Eigencurve can significantly outperform step decay in image classification tasks on CIFAR-10, especially when the number of epochs is small. Moreover, the theory inspires two simple learning rate schedulers for practical applications that can approximate eigencurve. For some problems, the optimal shape of the proposed schedulers resembles that of cosine decay, which sheds light to the success of cosine decay for such situations. For other situations, the proposed schedulers are superior to cosine decay.
With the rapid progress of generation technology, it has become necessary to attribute the origin of fake images. Existing works on fake image attribution perform multi-class classification on several Generative Adversarial Network (GAN) models and obtain high accuracies. While encouraging, these works are restricted to model-level attribution, only capable of handling images generated by seen models with a specific seed, loss and dataset, which is limited in real-world scenarios when fake images may be generated by privately trained models. This motivates us to ask whether it is possible to attribute fake images to the source models' architectures even if they are finetuned or retrained under different configurations. In this work, we present the first study on Deepfake Network Architecture Attribution to attribute fake images on architecture-level. Based on an observation that GAN architecture is likely to leave globally consistent fingerprints while traces left by model weights vary in different regions, we provide a simple yet effective solution named DNA-Det for this problem. Extensive experiments on multiple cross-test setups and a large-scale dataset demonstrate the effectiveness of DNA-Det.
Pre-trained vision-language models (e.g., CLIP) have shown promising zero-shot generalization in many downstream tasks with properly designed text prompts. Instead of relying on hand-engineered prompts, recent works learn prompts using the training data from downstream tasks. While effective, training on domain-specific data reduces a model's generalization capability to unseen new domains. In this work, we propose test-time prompt tuning (TPT), a method that can learn adaptive prompts on the fly with a single test sample. For image classification, TPT optimizes the prompt by minimizing the entropy with confidence selection so that the model has consistent predictions across different augmented views of each test sample. In evaluating generalization to natural distribution shifts, TPT improves the zero-shot top-1 accuracy of CLIP by 3.6% on average, surpassing previous prompt tuning approaches that require additional task-specific training data. In evaluating cross-dataset generalization with unseen categories, TPT performs on par with the state-of-the-art approaches that use additional training data. Project page: https://azshue.github.io/TPT.
The key to out-of-distribution detection is density estimation of the in-distribution data or of its feature representations. This is particularly challenging for dense anomaly detection in domains where the in-distribution data has a complex underlying structure. Nearest-Neighbors approaches have been shown to work well in object-centric data domains, such as industrial inspection and image classification. In this paper, we show that nearest-neighbor approaches also yield state-of-the-art results on dense novelty detection in complex driving scenes when working with an appropriate feature representation. In particular, we find that transformer-based architectures produce representations that yield much better similarity metrics for the task. We identify the multi-head structure of these models as one of the reasons, and demonstrate a way to transfer some of the improvements to CNNs. Ultimately, the approach is simple and non-invasive, i.e., it does not affect the primary segmentation performance, refrains from training on examples of anomalies, and achieves state-of-the-art results on RoadAnomaly, StreetHazards, and SegmentMeIfYouCan-Anomaly.
Vision Transformers convert images to sequences by slicing them into patches. The size of these patches controls a speed/accuracy tradeoff, with smaller patches leading to higher accuracy at greater computational cost, but changing the patch size typically requires retraining the model. In this paper, we demonstrate that simply randomizing the patch size at training time leads to a single set of weights that performs well across a wide range of patch sizes, making it possible to tailor the model to different compute budgets at deployment time. We extensively evaluate the resulting model, which we call FlexiViT, on a wide range of tasks, including classification, image-text retrieval, open-world detection, panoptic segmentation, and semantic segmentation, concluding that it usually matches, and sometimes outperforms, standard ViT models trained at a single patch size in an otherwise identical setup. Hence, FlexiViT training is a simple drop-in improvement for ViT that makes it easy to add compute-adaptive capabilities to most models relying on a ViT backbone architecture. Code and pre-trained models are available at https://github.com/google-research/big_vision
Neural Architecture Search (NAS) has been increasingly appealing to the society of object Re-Identification (ReID), for that task-specific architectures significantly improve the retrieval performance. Previous works explore new optimizing targets and search spaces for NAS ReID, yet they neglect the difference of training schemes between image classification and ReID. In this work, we propose a novel Twins Contrastive Mechanism (TCM) to provide more appropriate supervision for ReID architecture search. TCM reduces the category overlaps between the training and validation data, and assists NAS in simulating real-world ReID training schemes. We then design a Multi-Scale Interaction (MSI) search space to search for rational interaction operations between multi-scale features. In addition, we introduce a Spatial Alignment Module (SAM) to further enhance the attention consistency confronted with images from different sources. Under the proposed NAS scheme, a specific architecture is automatically searched, named as MSINet. Extensive experiments demonstrate that our method surpasses state-of-the-art ReID methods on both in-domain and cross-domain scenarios. Source code available in https://github.com/vimar-gu/MSINet.
Conventional training of deep neural networks requires a large number of the annotated image which is a laborious and time-consuming task, particularly for rare objects. Few-shot object detection (FSOD) methods offer a remedy by realizing robust object detection using only a few training samples per class. An unexplored challenge for FSOD is that instances from unlabeled novel classes that do not belong to the fixed set of training classes appear in the background. These objects behave similarly to label noise, leading to FSOD performance degradation. We develop a semi-supervised algorithm to detect and then utilize these unlabeled novel objects as positive samples during training to improve FSOD performance. Specifically, we propose a hierarchical ternary classification region proposal network (HTRPN) to localize the potential unlabeled novel objects and assign them new objectness labels. Our improved hierarchical sampling strategy for the region proposal network (RPN) also boosts the perception ability of the object detection model for large objects. Our experimental results indicate that our method is effective and outperforms the existing state-of-the-art (SOTA) FSOD methods.
Weight decay is a broadly used technique for training state-of-the-art deep networks from image classification to large language models. Despite its widespread usage and being extensively studied in the classical literature, its role remains poorly understood for deep learning. In this work, we highlight that the role of weight decay in modern deep learning is different from its regularization effect studied in classical learning theory. For deep networks on vision tasks trained with multipass SGD, we show how weight decay modifies the optimization dynamics enhancing the ever-present implicit regularization of SGD via the loss stabilization mechanism. In contrast, for large language models trained with nearly one-epoch training, we describe how weight decay balances the bias-variance tradeoff in stochastic optimization leading to lower training loss and improved training stability. Overall, we present a unifying perspective from ResNets on vision tasks to LLMs: weight decay is never useful as an explicit regularizer but instead changes the training dynamics in a desirable way. The code is available at https://github.com/tml-epfl/why-weight-decay
Downsampling operators break the shift invariance of convolutional neural networks (CNNs) and this affects the robustness of features learned by CNNs when dealing with even small pixel-level shift. Through a large-scale correlation analysis framework, we study shift invariance of CNNs by inspecting existing downsampling operators in terms of their maximum-sampling bias (MSB), and find that MSB is negatively correlated with shift invariance. Based on this crucial insight, we propose a learnable pooling operator called Translation Invariant Polyphase Sampling (TIPS) and two regularizations on the intermediate feature maps of TIPS to reduce MSB and learn translation-invariant representations. TIPS can be integrated into any CNN and can be trained end-to-end with marginal computational overhead. Our experiments demonstrate that TIPS results in consistent performance gains in terms of accuracy, shift consistency, and shift fidelity on multiple benchmarks for image classification and semantic segmentation compared to previous methods and also leads to improvements in adversarial and distributional robustness. TIPS results in the lowest MSB compared to all previous methods, thus explaining our strong empirical results.
신경-기호(Neural-Symbolic, NeSy) 시스템 분야는 빠르게 성장하고 있다. 제안된 접근법들은 신경망 방법과 기호 방법의 공생적 결합을 달성하는 데 큰 가능성을 보여준다. 그러나 일반적인 NeSy 모델링 패턴을 정리하고 범용 학습 접근법을 개발하기 위해서는 통합 프레임워크가 필요하다. 본 논문에서는 판별적(discriminative) 및 생성적(generative) NeSy 모델링을 위한 통합 수학적 프레임워크인 신경-기호 에너지 기반 모델(Neural-Symbolic Energy-Based Models, NeSy-EBMs)을 소개한다. 중요하게도, NeSy-EBMs는 주요 학습 손실(loss)의 그래디언트에 대한 일반적 표현식을 유도할 수 있게 해주며, 우리는 이중수준(bilevel) 최적화와 확률적 정책 최적화를 포함한 여러 분야의 방법을 활용하는 네 가지 학습 접근법 모음을 소개한다. 마지막으로, 확장성과 표현력을 갖추도록 설계된 오픈소스 NeSy-EBM 라이브러리인 신경 확률적 소프트 논리(Neural Probabilistic Soft Logic, NeuPSL)를 통해 NeSy-EBM 프레임워크를 구체화하여 NeSy 시스템의 실세계 적용을 용이하게 한다. 여러 데이터셋에 걸친 광범위한 경험적 분석을 통해, 우리는 이미지 분류, 그래프 노드 레이블링, 자율주행차 상황 인식, 질의응답을 포함한 과제들에서 NeuPSL을 통해 구체화된 NeSy-EBMs를 연구한다. 결과는 연구된 설정들에서 이러한 구체화의 실용적 유용성에 대한 증거를 제공한다.
Generating structured ASCII art using computational techniques demands a careful interplay between aesthetic representation and computational precision, requiring models that can effectively translate visual information into symbolic text characters. Although Convolutional Neural Networks (CNNs) have shown promise in this domain, the comparative performance of deep learning architectures and classical machine learning methods remains unexplored. This paper explores the application of contemporary ML and DL methods to generate structured ASCII art, focusing on three key criteria: fidelity, character classification accuracy, and output quality. We investigate deep learning architectures, including Multilayer Perceptrons (MLPs), ResNet, and MobileNetV2, alongside classical approaches such as Random Forests, Support Vector Machines (SVMs) and k-Nearest Neighbors (k-NN), trained on an augmented synthetic dataset of ASCII characters. Our results show that complex neural network architectures often fall short in producing high-quality ASCII art, whereas classical machine learning classifiers, despite their simplicity, achieve performance similar to CNNs. Our findings highlight the strength of classical methods in bridging model simplicity with output quality, offering new insights into ASCII art synthesis and machine learning on image data with low dimensionality.
The goal of this work is to characterize the representational impact that foveation operations have for machine vision systems, inspired by the foveated human visual system, which has higher acuity at the center of gaze and texture-like encoding in the periphery. To do so, we introduce models consisting of a first-stage \textit{fixed} image transform followed by a second-stage \textit{learnable} convolutional neural network, and we varied the first stage component. The primary model has a foveated-textural input stage, which we compare to a model with foveated-blurred input and a model with spatially-uniform blurred input (both matched for perceptual compression), and a final reference model with minimal input-based compression. We find that: 1) the foveated-texture model shows similar scene classification accuracy as the reference model despite its compressed input, with greater i.i.d. generalization than the other models; 2) the foveated-texture model has greater sensitivity to high-spatial frequency information and greater robustness to occlusion, w.r.t the comparison models; 3) both the foveated systems, show a stronger center image-bias relative to the spatially-uniform systems even with a weight sharing constraint. Critically, these results are preserved over different classical CNN architectures throughout their learning dynamics. Altogether, this suggests that foveation with peripheral texture-based computations yields an efficient, distinct, and robust representational format of scene information, and provides symbiotic computational insight into the representational consequences that texture-based peripheral encoding may have for processing in the human visual system, while also potentially inspiring the next generation of computer vision models via spatially-adaptive computation. Code + Data available here: https://github.com/ArturoDeza/EmergentProperties