Recently, MLP-based vision backbones have achieved promising performance in several visual recognition tasks. However, the existing MLP-based methods directly aggregate tokens with static weights, leaving the adaptability to different images untouched. Moreover, Recent research demonstrates that MLP-Transformer is great at creating long-range dependencies but ineffective at catching high frequencies that primarily transmit local information, which prevents it from applying to the downstream dense prediction tasks, such as semantic segmentation. To address these challenges, we propose a content-adaptive yet computationally efficient structure, dubbed Dynamic Spectrum Mixer (DSM). The DSM represents token interactions in the frequency domain by employing the Discrete Cosine Transform, which can learn long-term spatial dependencies with log-linear complexity. Furthermore, a dynamic spectrum weight generation layer is proposed as the spectrum bands selector, which could emphasize the informative frequency bands while diminishing others. To this end, the technique can efficiently learn detailed features from visual input that contains both high- and low-frequency information. Extensive experiments show that DSM is a powerful and adaptable backbone for a range of visual recognition tasks. Particularly, DSM outperforms previous transformer-based and MLP-based models, on image classification, object detection, and semantic segmentation tasks, such as 83.8 \% top-1 accuracy on ImageNet, and 49.9 \% mIoU on ADE20K.
Convolution is a fundamental operation in image processing and machine learning. Aimed primarily at maintaining image size, padding is a key ingredient of convolution, which, however, can introduce undesirable boundary effects. We present a non-padding-based method for size-keeping convolution based on the preservation of differential characteristics of kernels. The main idea is to make convolution over an incomplete sliding window "collapse" to a linear differential operator evaluated locally at its central pixel, which no longer requires information from the neighbouring missing pixels. While the underlying theory is rigorous, our final formula turns out to be simple: the convolution over an incomplete window is achieved by convolving its nearest complete window with a transformed kernel. This formula is computationally lightweight, involving neither interpolation or extrapolation nor restrictions on image and kernel sizes. Our method favours data with smooth boundaries, such as high-resolution images and fields from physics. Our experiments include: i) filtering analytical and non-analytical fields from computational physics and, ii) training convolutional neural networks (CNNs) for the tasks of image classification, semantic segmentation and super-resolution reconstruction. In all these experiments, our method has exhibited visible superiority over the compared ones.
British Antarctic Survey (BAS) researchers launch annual expeditions to the Antarctic in order to estimate Antarctic Krill biomass and assess the change from previous years. These comparisons provide insight into the effects of the current environment on this key component of the marine food chain. In this work we have developed tools for automating the data collection and analysis process, using web-based image annotation tools and deep learning image classification and regression models. We achieve highly accurate krill instance segmentation results with an average 77.28% AP score, as well as separate maturity stage and length estimation of krill specimens with 62.99% accuracy and a 1.98mm length error respectively.
Vision-language models (VLMs) pre-trained on web-scale datasets have demonstrated remarkable capabilities on downstream tasks when fine-tuned with minimal data. However, many VLMs rely on proprietary data and are not open-source, which restricts the use of white-box approaches for fine-tuning. As such, we aim to develop a black-box approach to optimize VLMs through natural language prompts, thereby avoiding the need to access model parameters, feature embeddings, or even output logits. We propose employing chat-based LLMs to search for the best text prompt for VLMs. Specifically, we adopt an automatic hill-climbing procedure that converges to an effective prompt by evaluating the performance of current prompts and asking LLMs to refine them based on textual feedback, all within a conversational process without human-in-the-loop. In a challenging 1-shot image classification setup, our simple approach surpasses the white-box continuous prompting method (CoOp) by an average of 1.5% across 11 datasets including ImageNet. Our approach also outperforms both human-engineered and LLM-generated prompts. We highlight the advantage of conversational feedback that incorporates both positive and negative prompts, suggesting that LLMs can utilize the implicit gradient direction in textual feedback for a more efficient search. In addition, we find that the text prompts generated through our strategy are not only more interpretable but also transfer well across different VLM architectures in a black-box manner. Lastly, we apply our framework to optimize the state-of-the-art black-box VLM (DALL-E 3) for text-to-image generation, prompt inversion, and personalization.
Deep neural networks (DNNs) are increasingly proposed as models of human vision, bolstered by their impressive performance on image classification and object recognition tasks. Yet, the extent to which DNNs capture fundamental aspects of human vision such as color perception remains unclear. Here, we develop novel experiments for evaluating the perceptual coherence of color embeddings in DNNs, and we assess how well these algorithms predict human color similarity judgments collected via an online survey. We find that state-of-the-art DNN architectures $-$ including convolutional neural networks and vision transformers $-$ provide color similarity judgments that strikingly diverge from human color judgments of (i) images with controlled color properties, (ii) images generated from online searches, and (iii) real-world images from the canonical CIFAR-10 dataset. We compare DNN performance against an interpretable and cognitively plausible model of color perception based on wavelet decomposition, inspired by foundational theories in computational neuroscience. While one deep learning model $-$ a convolutional DNN trained on a style transfer task $-$ captures some aspects of human color perception, our wavelet algorithm provides more coherent color embeddings that better predict human color judgments compared to all DNNs we examine. These results hold when altering the high-level visual task used to train similar DNN architectures (e.g., image classification versus image segmentation), as well as when examining the color embeddings of different layers in a given DNN architecture. These findings break new ground in the effort to analyze the perceptual representations of machine learning algorithms and to improve their ability to serve as cognitively plausible models of human vision. Implications for machine learning, human perception, and embodied cognition are discussed.
Visual document understanding (VDU) has rapidly advanced with the development of powerful multi-modal language models. However, these models typically require extensive document pre-training data to learn intermediate representations and often suffer a significant performance drop in real-world online industrial settings. A primary issue is their heavy reliance on OCR engines to extract local positional information within document pages, which limits the models' ability to capture global information and hinders their generalizability, flexibility, and robustness. In this paper, we introduce GlobalDoc, a cross-modal transformer-based architecture pre-trained in a self-supervised manner using three novel pretext objective tasks. GlobalDoc improves the learning of richer semantic concepts by unifying language and visual representations, resulting in more transferable models. For proper evaluation, we also propose two novel document-level downstream VDU tasks, Few-Shot Document Image Classification (DIC) and Content-based Document Image Retrieval (DIR), designed to simulate industrial scenarios more closely. Extensive experimentation has been conducted to demonstrate GlobalDoc's effectiveness in practical settings.
Biological synaptic transmission is unreliable, and this unreliability likely degrades neural circuit performance. While there are biophysical mechanisms that can increase reliability, for instance by increasing vesicle release probability, these mechanisms cost energy. We examined four such mechanisms along with the associated scaling of the energetic costs. We then embedded these energetic costs for reliability in artificial neural networks (ANN) with trainable stochastic synapses, and trained these networks on standard image classification tasks. The resulting networks revealed a tradeoff between circuit performance and the energetic cost of synaptic reliability. Additionally, the optimised networks exhibited two testable predictions consistent with pre-existing experimental data. Specifically, synapses with lower variability tended to have 1) higher input firing rates and 2) lower learning rates. Surprisingly, these predictions also arise when synapse statistics are inferred through Bayesian inference. Indeed, we were able to find a formal, theoretical link between the performance-reliability cost tradeoff and Bayesian inference. This connection suggests two incompatible possibilities: evolution may have chanced upon a scheme for implementing Bayesian inference by optimising energy efficiency, or alternatively, energy efficient synapses may display signatures of Bayesian inference without actually using Bayes to reason about uncertainty.
This paper introduces a novel approach to enhance the performance of the stochastic gradient descent (SGD) algorithm by incorporating a modified decay step size based on $\frac{1}{\sqrt{t}}$. The proposed step size integrates a logarithmic term, leading to the selection of smaller values in the final iterations. Our analysis establishes a convergence rate of $O(\frac{\ln T}{\sqrt{T}})$ for smooth non-convex functions without the Polyak-{\L}ojasiewicz condition. To evaluate the effectiveness of our approach, we conducted numerical experiments on image classification tasks using the FashionMNIST, and CIFAR10 datasets, and the results demonstrate significant improvements in accuracy, with enhancements of $0.5\%$ and $1.4\%$ observed, respectively, compared to the traditional $\frac{1}{\sqrt{t}}$ step size. The source code can be found at \\\url{https://github.com/Shamaeem/LNSQRTStepSize}.
We present an effective application of quantum machine learning in histopathological cancer detection. The study here emphasizes two primary applications of hybrid classical-quantum Deep Learning models. The first application is to build a classification model for histopathological cancer detection using the quantum transfer learning strategy. The second application is to test the performance of this model for various adversarial attacks. Rather than using a single transfer learning model, the hybrid classical-quantum models are tested using multiple transfer learning models, especially ResNet18, VGG-16, Inception-v3, and AlexNet as feature extractors and integrate it with several quantum circuit-based variational quantum circuits (VQC) with high expressibility. As a result, we provide a comparative analysis of classical models and hybrid classical-quantum transfer learning models for histopathological cancer detection under several adversarial attacks. We compared the performance accuracy of the classical model with the hybrid classical-quantum model using pennylane default quantum simulator. We also observed that for histopathological cancer detection under several adversarial attacks, Hybrid Classical-Quantum (HCQ) models provided better accuracy than classical image classification models.
We propose a simple, efficient, and accurate method for detecting out-of-distribution (OOD) data for trained neural networks. We propose a novel descriptor, HAct - activation histograms, for OOD detection, that is, probability distributions (approximated by histograms) of output values of neural network layers under the influence of incoming data. We formulate an OOD detector based on HAct descriptors. We demonstrate that HAct is significantly more accurate than state-of-the-art in OOD detection on multiple image classification benchmarks. For instance, our approach achieves a true positive rate (TPR) of 95% with only 0.03% false-positives using Resnet-50 on standard OOD benchmarks, outperforming previous state-of-the-art by 20.67% in the false positive rate (at the same TPR of 95%). The computational efficiency and the ease of implementation makes HAct suitable for online implementation in monitoring deployed neural networks in practice at scale.
Repeated parameter sharing in federated learning causes significant information leakage about private data, thus defeating its main purpose: data privacy. Mitigating the risk of this information leakage, using state of the art differentially private algorithms, also does not come for free. Randomized mechanisms can prevent convergence of models on learning even the useful representation functions, especially if there is more disagreement between local models on the classification functions (due to data heterogeneity). In this paper, we consider a representation federated learning objective that encourages various parties to collaboratively refine the consensus part of the model, with differential privacy guarantees, while separately allowing sufficient freedom for local personalization (without releasing it). We prove that in the linear representation setting, while the objective is non-convex, our proposed new algorithm \DPFEDREP\ converges to a ball centered around the \emph{global optimal} solution at a linear rate, and the radius of the ball is proportional to the reciprocal of the privacy budget. With this novel utility analysis, we improve the SOTA utility-privacy trade-off for this problem by a factor of $\sqrt{d}$, where $d$ is the input dimension. We empirically evaluate our method with the image classification task on CIFAR10, CIFAR100, and EMNIST, and observe a significant performance improvement over the prior work under the same small privacy budget. The code can be found in this link: https://github.com/shenzebang/CENTAUR-Privacy-Federated-Representation-Learning.
Due to the widespread availability of data, machine learning (ML) algorithms are increasingly being implemented in distributed topologies, wherein various nodes collaborate to train ML models via the coordination of a central server. However, distributed learning approaches face significant vulnerabilities, primarily stemming from two potential threats. Firstly, the presence of Byzantine nodes poses a risk of corrupting the learning process by transmitting inaccurate information to the server. Secondly, a curious server may compromise the privacy of individual nodes, sometimes reconstructing the entirety of the nodes' data. Homomorphic encryption (HE) has emerged as a leading security measure to preserve privacy in distributed learning under non-Byzantine scenarios. However, the extensive computational demands of HE, particularly for high-dimensional ML models, have deterred attempts to design purely homomorphic operators for non-linear robust aggregators. This paper introduces SABLE, the first homomorphic and Byzantine robust distributed learning algorithm. SABLE leverages HTS, a novel and efficient homomorphic operator implementing the prominent coordinate-wise trimmed mean robust aggregator. Designing HTS enables us to implement HMED, a novel homomorphic median aggregator. Extensive experiments on standard ML tasks demonstrate that SABLE achieves practical execution times while maintaining an ML accuracy comparable to its non-private counterpart.
Continual learning is a challenging problem in which models need to be trained on non-stationary data across sequential tasks for class-incremental learning. While previous methods have focused on using either regularization or rehearsal-based frameworks to alleviate catastrophic forgetting in image classification, they are limited to a single modality and cannot learn compact class-aware cross-modal representations for continual audio-visual learning. To address this gap, we propose a novel class-incremental grouping network (CIGN) that can learn category-wise semantic features to achieve continual audio-visual learning. Our CIGN leverages learnable audio-visual class tokens and audio-visual grouping to continually aggregate class-aware features. Additionally, it utilizes class tokens distillation and continual grouping to prevent forgetting parameters learned from previous tasks, thereby improving the model's ability to capture discriminative audio-visual categories. We conduct extensive experiments on VGGSound-Instruments, VGGSound-100, and VGG-Sound Sources benchmarks. Our experimental results demonstrate that the CIGN achieves state-of-the-art audio-visual class-incremental learning performance. Code is available at https://github.com/stoneMo/CIGN.
Automated detection of Gallbladder Cancer (GBC) from Ultrasound (US) images is an important problem, which has drawn increased interest from researchers. However, most of these works use difficult-to-acquire information such as bounding box annotations or additional US videos. In this paper, we focus on GBC detection using only image-level labels. Such annotation is usually available based on the diagnostic report of a patient, and do not require additional annotation effort from the physicians. However, our analysis reveals that it is difficult to train a standard image classification model for GBC detection. This is due to the low inter-class variance (a malignant region usually occupies only a small portion of a US image), high intra-class variance (due to the US sensor capturing a 2D slice of a 3D object leading to large viewpoint variations), and low training data availability. We posit that even when we have only the image level label, still formulating the problem as object detection (with bounding box output) helps a deep neural network (DNN) model focus on the relevant region of interest. Since no bounding box annotations is available for training, we pose the problem as weakly supervised object detection (WSOD). Motivated by the recent success of transformer models in object detection, we train one such model, DETR, using multi-instance-learning (MIL) with self-supervised instance selection to suit the WSOD task. Our proposed method demonstrates an improvement of AP and detection sensitivity over the SOTA transformer-based and CNN-based WSOD methods. Project page is at https://gbc-iitd.github.io/wsod-gbc
Data augmentation (DA) is widely used to improve the generalization of neural networks by enforcing the invariances and symmetries to pre-defined transformations applied to input data. However, a fixed augmentation policy may have different effects on each sample in different training stages but existing approaches cannot adjust the policy to be adaptive to each sample and the training model. In this paper, we propose Model Adaptive Data Augmentation (MADAug) that jointly trains an augmentation policy network to teach the model when to learn what. Unlike previous work, MADAug selects augmentation operators for each input image by a model-adaptive policy varying between training stages, producing a data augmentation curriculum optimized for better generalization. In MADAug, we train the policy through a bi-level optimization scheme, which aims to minimize a validation-set loss of a model trained using the policy-produced data augmentations. We conduct an extensive evaluation of MADAug on multiple image classification tasks and network architectures with thorough comparisons to existing DA approaches. MADAug outperforms or is on par with other baselines and exhibits better fairness: it brings improvement to all classes and more to the difficult ones. Moreover, MADAug learned policy shows better performance when transferred to fine-grained datasets. In addition, the auto-optimized policy in MADAug gradually introduces increasing perturbations and naturally forms an easy-to-hard curriculum.
eXplanation Based Learning (XBL) is an interactive learning approach that provides a transparent method of training deep learning models by interacting with their explanations. XBL augments loss functions to penalize a model based on deviation of its explanations from user annotation of image features. The literature on XBL mostly depends on the intersection of visual model explanations and image feature annotations. We present a method to add a distance-aware explanation loss to categorical losses that trains a learner to focus on important regions of a training dataset. Distance is an appropriate approach for calculating explanation loss since visual model explanations such as Gradient-weighted Class Activation Mapping (Grad-CAMs) are not strictly bounded as annotations and their intersections may not provide complete information on the deviation of a model's focus from relevant image regions. In addition to assessing our model using existing metrics, we propose an interpretability metric for evaluating visual feature-attribution based model explanations that is more informative of the model's performance than existing metrics. We demonstrate performance of our proposed method on three image classification tasks.
Tsetlin Machines (TMs) provide a fundamental shift from arithmetic-based to logic-based machine learning. Supporting convolution, they deal successfully with image classification datasets like MNIST, Fashion-MNIST, and CIFAR-2. However, the TM struggles with getting state-of-the-art performance on CIFAR-10 and CIFAR-100, representing more complex tasks. This paper introduces plug-and-play collaboration between specialized TMs, referred to as TM Composites. The collaboration relies on a TM's ability to specialize during learning and to assess its competence during inference. When teaming up, the most confident TMs make the decisions, relieving the uncertain ones. In this manner, a TM Composite becomes more competent than its members, benefiting from their specializations. The collaboration is plug-and-play in that members can be combined in any way, at any time, without fine-tuning. We implement three TM specializations in our empirical evaluation: Histogram of Gradients, Adaptive Gaussian Thresholding, and Color Thermometers. The resulting TM Composite increases accuracy on Fashion-MNIST by two percentage points, CIFAR-10 by twelve points, and CIFAR-100 by nine points, yielding new state-of-the-art results for TMs. Overall, we envision that TM Composites will enable an ultra-low energy and transparent alternative to state-of-the-art deep learning on more tasks and datasets.
Conformal prediction (CP) generates a set of predictions for a given test sample such that the prediction set almost always contains the true label (e.g., 99.5\% of the time). CP provides comprehensive predictions on possible labels of a given test sample, and the size of the set indicates how certain the predictions are (e.g., a set larger than one is `uncertain'). Such distinct properties of CP enable effective collaborations between human experts and medical AI models, allowing efficient intervention and quality check in clinical decision-making. In this paper, we propose a new method called Reliable-Region-Based Conformal Prediction (RR-CP), which aims to impose a stronger statistical guarantee so that the user-specified error rate (e.g., 0.5\%) can be achieved in the test time, and under this constraint, the size of the prediction set is optimized (to be small). We consider a small prediction set size an important measure only when the user-specified error rate is achieved. Experiments on five public datasets show that our RR-CP performs well: with a reasonably small-sized prediction set, it achieves the user-specified error rate (e.g., 0.5\%) significantly more frequently than exiting CP methods.
Gradient inversion attacks are an ubiquitous threat in federated learning as they exploit gradient leakage to reconstruct supposedly private training data. Recent work has proposed to prevent gradient leakage without loss of model utility by incorporating a PRivacy EnhanCing mODulE (PRECODE) based on variational modeling. Without further analysis, it was shown that PRECODE successfully protects against gradient inversion attacks. In this paper, we make multiple contributions. First, we investigate the effect of PRECODE on gradient inversion attacks to reveal its underlying working principle. We show that variational modeling introduces stochasticity into the gradients of PRECODE and the subsequent layers in a neural network. The stochastic gradients of these layers prevent iterative gradient inversion attacks from converging. Second, we formulate an attack that disables the privacy preserving effect of PRECODE by purposefully omitting stochastic gradients during attack optimization. To preserve the privacy preserving effect of PRECODE, our analysis reveals that variational modeling must be placed early in the network. However, early placement of PRECODE is typically not feasible due to reduced model utility and the exploding number of additional model parameters. Therefore, as a third contribution, we propose a novel privacy module -- the Convolutional Variational Bottleneck (CVB) -- that can be placed early in a neural network without suffering from these drawbacks. We conduct an extensive empirical study on three seminal model architectures and six image classification datasets. We find that all architectures are susceptible to gradient leakage attacks, which can be prevented by our proposed CVB. Compared to PRECODE, we show that our novel privacy module requires fewer trainable parameters, and thus computational and communication costs, to effectively preserve privacy.
Advancements in computer vision research have put transformer architecture as the state of the art in computer vision tasks. One of the known drawbacks of the transformer architecture is the high number of parameters, this can lead to a more complex and inefficient algorithm. This paper aims to reduce the number of parameters and in turn, made the transformer more efficient. We present Sparse Transformer (SparTa) Block, a modified transformer block with an addition of a sparse token converter that reduces the number of tokens used. We use the SparTa Block inside the Swin T architecture (SparseSwin) to leverage Swin capability to downsample its input and reduce the number of initial tokens to be calculated. The proposed SparseSwin model outperforms other state of the art models in image classification with an accuracy of 86.96%, 97.43%, and 85.35% on the ImageNet100, CIFAR10, and CIFAR100 datasets respectively. Despite its fewer parameters, the result highlights the potential of a transformer architecture using a sparse token converter with a limited number of tokens to optimize the use of the transformer and improve its performance.