Leveraging transfer learning has recently been shown to be an effective strategy for training large models with Differential Privacy (DP). Moreover, somewhat surprisingly, recent works have found that privately training just the last layer of a pre-trained model provides the best utility with DP. While past studies largely rely on algorithms like DP-SGD for training large models, in the specific case of privately learning from features, we observe that computational burden is low enough to allow for more sophisticated optimization schemes, including second-order methods. To that end, we systematically explore the effect of design parameters such as loss function and optimization algorithm. We find that, while commonly used logistic regression performs better than linear regression in the non-private setting, the situation is reversed in the private setting. We find that linear regression is much more effective than logistic regression from both privacy and computational aspects, especially at stricter epsilon values ($\epsilon < 1$). On the optimization side, we also explore using Newton's method, and find that second-order information is quite helpful even with privacy, although the benefit significantly diminishes with stricter privacy guarantees. While both methods use second-order information, least squares is effective at lower epsilons while Newton's method is effective at larger epsilon values. To combine the benefits of both, we propose a novel algorithm called DP-FC, which leverages feature covariance instead of the Hessian of the logistic regression loss and performs well across all $\epsilon$ values we tried. With this, we obtain new SOTA results on ImageNet-1k, CIFAR-100 and CIFAR-10 across all values of $\epsilon$ typically considered. Most remarkably, on ImageNet-1K, we obtain top-1 accuracy of 88\% under (8, $8 * 10^{-7}$)-DP and 84.3\% under (0.1, $8 * 10^{-7}$)-DP.
A common challenge in real world classification scenarios with sequentially appending target domain data is insufficient training datasets during the training phase. Therefore, conventional deep learning and transfer learning classifiers are not applicable especially when individual classes are not represented or are severely underrepresented at the outset. In many technical domains, however, it is only the defect or worn reject classes that are insufficiently represented, while the non-defect class is often available from the beginning. The proposed classification approach addresses such conditions and is based on a CNN encoder. Following a contrastive learning approach, it is trained with a modified triplet loss function using two datasets: Besides the non-defective target domain class 1st dataset, a state-of-the-art labeled source domain dataset that contains highly related classes e.g., a related manufacturing error or wear defect but originates from a highly different domain e.g., different product, material, or appearance = 2nd dataset is utilized. The approach learns the classification features from the source domain dataset while at the same time learning the differences between the source and the target domain in a single training step, aiming to transfer the relevant features to the target domain. The classifier becomes sensitive to the classification features and by architecture robust against the highly domain-specific context. The approach is benchmarked in a technical and a non-technical domain and shows convincing classification results. In particular, it is shown that the domain generalization capabilities and classification results are improved by the proposed architecture, allowing for larger domain shifts between source and target domains.
Due to the rapid temporal and fine-grained nature of complex human assembly atomic actions, traditional action segmentation approaches requiring the spatial (and often temporal) down sampling of video frames often loose vital fine-grained spatial and temporal information required for accurate classification within the manufacturing domain. In order to fully utilise higher resolution video data (often collected within the manufacturing domain) and facilitate real time accurate action segmentation - required for human robot collaboration - we present a novel hand location guided high resolution feature enhanced model. We also propose a simple yet effective method of deploying offline trained action recognition models for real time action segmentation on temporally short fine-grained actions, through the use of surround sampling while training and temporally aware label cleaning at inference. We evaluate our model on a novel action segmentation dataset containing 24 (+background) atomic actions from video data of a real world robotics assembly production line. Showing both high resolution hand features as well as traditional frame wide features improve fine-grained atomic action classification, and that though temporally aware label clearing our model is capable of surpassing similar encoder/decoder methods, while allowing for real time classification.
Standard semantic segmentation models owe their success to curated datasets with a fixed set of semantic categories, without contemplating the possibility of identifying unknown objects from novel categories. Existing methods in outlier detection suffer from a lack of smoothness and objectness in their predictions, due to limitations of the per-pixel classification paradigm. Furthermore, additional training for detecting outliers harms the performance of known classes. In this paper, we explore another paradigm with region-level classification to better segment unknown objects. We show that the object queries in mask classification tend to behave like one \vs all classifiers. Based on this finding, we propose a novel outlier scoring function called RbA by defining the event of being an outlier as being rejected by all known classes. Our extensive experiments show that mask classification improves the performance of the existing outlier detection methods, and the best results are achieved with the proposed RbA. We also propose an objective to optimize RbA using minimal outlier supervision. Further fine-tuning with outliers improves the unknown performance, and unlike previous methods, it does not degrade the inlier performance.
We present XKD, a novel self-supervised framework to learn meaningful representations from unlabelled videos. XKD is trained with two pseudo objectives. First, masked data reconstruction is performed to learn modality-specific representations from audio and visual streams. Next, self-supervised cross-modal knowledge distillation is performed between the two modalities through a teacher-student setup to learn complementary information. We introduce a novel domain alignment strategy to tackle domain discrepancy between audio and visual modalities enabling effective cross-modal knowledge distillation. Additionally, to develop a general-purpose network capable of handling both audio and visual streams, modality-agnostic variants of XKD are introduced, which use the same pretrained backbone for different audio and visual tasks. Our proposed cross-modal knowledge distillation improves video action classification by $8\%$ to $14\%$ on UCF101, HMDB51, and Kinetics400. Additionally, XKD improves multimodal action classification by $5.5\%$ on Kinetics-Sound. XKD shows state-of-the-art performance in sound classification on ESC50, achieving top-1 accuracy of $96.5\%$.
One of the central motivations behind Natural Language Processing is detecting patterns. Given a text document, the task of identifying the context is known to be as topic classification. This paper explores the performance of three different classifiers namely Decision Tree, K-Nearest Neighbors, and Multinomial Naive Bayes on a topic classification task (with six topic classes). The evaluation is done on the basis of accuracy, precision, recall, and f1-score based results. Among those three aforementioned classifiers, we have selected the Multinomial Naïve Bayes a sour best model using which we achieved 91.8% accuracy.
In this paper, we propose a dictionary screening method for embedding compression in text classification tasks. The key purpose of this method is to evaluate the importance of each keyword in the dictionary. To this end, we first train a pre-specified recurrent neural network-based model using a full dictionary. This leads to a benchmark model, which we then use to obtain the predicted class probabilities for each sample in a dataset. Next, to evaluate the impact of each keyword in affecting the predicted class probabilities, we develop a novel method for assessing the importance of each keyword in a dictionary. Consequently, each keyword can be screened, and only the most important keywords are reserved. With these screened keywords, a new dictionary with a considerably reduced size can be constructed. Accordingly, the original text sequence can be substantially compressed. The proposed method leads to significant reductions in terms of parameters, average text sequence, and dictionary size. Meanwhile, the prediction power remains very competitive compared to the benchmark model. Extensive numerical studies are presented to demonstrate the empirical performance of the proposed method.
We propose a novel image dataset focused on tiny faces wearing face masks for mask classification purposes, dubbed Small Face MASK (SF-MASK), composed of a collection made from 20k low-resolution images exported from diverse and heterogeneous datasets, ranging from 7 x 7 to 64 x 64 pixel resolution. An accurate visualization of this collection, through counting grids, made it possible to highlight gaps in the variety of poses assumed by the heads of the pedestrians. In particular, faces filmed by very high cameras, in which the facial features appear strongly skewed, are absent. To address this structural deficiency, we produced a set of synthetic images which resulted in a satisfactory covering of the intra-class variance. Furthermore, a small subsample of 1701 images contains badly worn face masks, opening to multi-class classification challenges. Experiments on SF-MASK focus on face mask classification using several classifiers. Results show that the richness of SF-MASK (real + synthetic images) leads all of the tested classifiers to perform better than exploiting comparative face mask datasets, on a fixed 1077 images testing set. Dataset and evaluation code are publicly available here: https://github.com/HumaticsLAB/sf-mask
Recently, the vision transformer (ViT) has made breakthroughs in image recognition. Its self-attention mechanism (MSA) can extract discriminative labeling information of different pixel blocks to improve image classification accuracy. However, the classification marks in their deep layers tend to ignore local features between layers. In addition, the embedding layer will be fixed-size pixel blocks. Input network Inevitably introduces additional image noise. To this end, we study a data augmentation vision transformer (DAVT) based on data augmentation and proposes a data augmentation method for attention cropping, which uses attention weights as the guide to crop images and improve the ability of the network to learn critical features. Secondly, we also propose a hierarchical attention selection (HAS) method, which improves the ability of discriminative markers between levels of learning by filtering and fusing labels between levels. Experimental results show that the accuracy of this method on the two general datasets, CUB-200-2011, and Stanford Dogs, is better than the existing mainstream methods, and its accuracy is 1.4\% and 1.6\% higher than the original ViT, respectively
Android malware attacks are increasing daily at a tremendous volume, making Android users more vulnerable to cyber-attacks. Researchers have developed many machine learning (ML)/ deep learning (DL) techniques to detect and mitigate android malware attacks. However, due to technological advancement, there is a rise in android mobile devices. Furthermore, the devices are geographically dispersed, resulting in distributed data. In such scenario, traditional ML/DL techniques are infeasible since all of these approaches require the data to be kept in a central system; this may provide a problem for user privacy because of the massive proliferation of Android mobile devices; putting the data in a central system creates an overhead. Also, the traditional ML/DL-based android malware classification techniques are not scalable. Researchers have proposed federated learning (FL) based android malware classification system to solve the privacy preservation and scalability with high classification performance. In traditional FL, Federated Averaging (FedAvg) is utilized to construct the global model at each round by merging all of the local models obtained from all of the customers that participated in the FL. However, the conventional FedAvg has a disadvantage: if one poor-performing local model is included in global model development for each round, it may result in an under-performing global model. Because FedAvg favors all local models equally when averaging. To address this issue, our main objective in this work is to design a dynamic weighted federated averaging (DW-FedAvg) strategy in which the weights for each local model are automatically updated based on their performance at the client. The DW-FedAvg is evaluated using four popular benchmark datasets, Melgenome, Drebin, Kronodroid and Tuandromd used in android malware classification research.
Lensless imaging can provide visual privacy due to the highly multiplexed characteristic of its measurements. However, this alone is a weak form of security, as various adversarial attacks can be designed to invert the one-to-many scene mapping of such cameras. In this work, we enhance the privacy provided by lensless imaging by (1) downsampling at the sensor and (2) using a programmable mask with variable patterns as our optical encoder. We build a prototype from a low-cost LCD and Raspberry Pi components, for a total cost of around 100 USD. This very low price point allows our system to be deployed and leveraged in a broad range of applications. In our experiments, we first demonstrate the viability and reconfigurability of our system by applying it to various classification tasks: MNIST, CelebA (face attributes), and CIFAR10. By jointly optimizing the mask pattern and a digital classifier in an end-to-end fashion, low-dimensional, privacy-enhancing embeddings are learned directly at the sensor. Secondly, we show how the proposed system, through variable mask patterns, can thwart adversaries that attempt to invert the system (1) via plaintext attacks or (2) in the event of camera parameters leaks. We demonstrate the defense of our system to both risks, with 55% and 26% drops in image quality metrics for attacks based on model-based convex optimization and generative neural networks respectively. We open-source a wave propagation and camera simulator needed for end-to-end optimization, the training software, and a library for interfacing with the camera.
We present a hybrid classical-quantum approach to the binary classification of polymer structures. Two polymer classes visual (VIS) and near-infrared (NIR) are defined based on the size of the polymer gaps. The hybrid approach combines one of the three methods, Gaussian Kernel Method, Quantum-Enhanced Random Kitchen Sinks or Variational Quantum Classifier, implemented by linear quantum photonic circuits (LQPCs), with a classical deep neural network (DNN) feature extractor. The latter extracts from the classical data information about samples chemical structure. It also reduces the data dimensions yielding compact 2-dimensional data vectors that are then fed to the LQPCs. We adopt the photonic-based data-embedding scheme, proposed by Gan et al. [EPJ Quantum Technol. 9, 16 (2022)] to embed the classical 2-dimensional data vectors into the higher-dimensional Fock space. This hybrid classical-quantum strategy permits to obtain accurate noisy intermediate-scale quantum-compatible classifiers by leveraging Fock states with only a few photons. The models obtained using either of the three hybrid methods successfully classified the VIS and NIR polymers. Their accuracy is comparable as measured by their scores ranging from 0.86 to 0.88. These findings demonstrate that our hybrid approach that uses photonic quantum computing captures chemistry and structure-property correlation patterns in real polymer data. They also open up perspectives of employing quantum computing to complex chemical structures when a larger number of logical qubits is available.
Skin cancer is a fatal disease that takes a heavy toll over human lives annually. The colored skin images show a significant degree of resemblance between different skin lesions such as melanoma and nevus, making identification and diagnosis more challenging. Melanocytic nevi may mature to cause fatal melanoma. Therefore, the current management protocol involves the removal of those nevi that appear intimidating. However, this necessitates resilient classification paradigms for classifying benign and malignant melanocytic nevi. Early diagnosis necessitates a dependable automated system for melanocytic nevi classification to render diagnosis efficient, timely, and successful. An automated classification algorithm is proposed in the given research. A neural network previously-trained on a separate problem statement is leveraged in this technique for classifying melanocytic nevus images. The suggested method uses BigTransfer (BiT), a ResNet-based transfer learning approach for classifying melanocytic nevi as malignant or benign. The results obtained are compared to that of current techniques, and the new method's classification rate is proven to outperform that of existing methods.
Recent studies have applied deep learning methods such as convolutional recurrent neural networks (CRNs) and Transformers to brain disease classification based on dynamic functional connectivity networks (dFCNs), such as Alzheimer's disease (AD), achieving better performance than traditional machine learning methods. However, in CRNs, the continuous convolution operations used to obtain high-order aggregation features may overlook the non-linear correlation between different brain regions due to the essence of convolution being the linear weighted sum of local elements. Inspired by modern neuroscience on the research of covert attention in the nervous system, we introduce the self-attention mechanism, a core module of Transformers, to model diversified covert attention patterns and apply these patterns to reconstruct high-order sequence features of dFCNs in order to learn complex dynamic changes in brain information flow. Therefore, we propose a novel CRN method based on diversified covert attention patterns, DCA-CRN, which combines the advantages of CRNs in capturing local spatio-temporal features and sequence change patterns, as well as Transformers in learning global and high-order correlation features. Experimental results on the ADNI and ADHD-200 datasets demonstrate the prediction performance and generalization ability of our proposed method.
Bidirectional Encoder Representations from Transformers or BERT~\cite{devlin-etal-2019-bert} has been one of the base models for various NLP tasks due to its remarkable performance. Variants customized for different languages and tasks are proposed to further improve the performance. In this work, we investigate supervised continued pre-training~\cite{gururangan-etal-2020-dont} on BERT for Chinese topic classification task. Specifically, we incorporate prompt-based learning and contrastive learning into the pre-training. To adapt to the task of Chinese topic classification, we collect around 2.1M Chinese data spanning various topics. The pre-trained Chinese Topic Classification BERTs (TCBERTs) with different parameter sizes are open-sourced at \url{https://huggingface.co/IDEA-CCNL}.
Multi-label classification (MLC) is an ML task of predictive modeling in which a data instance can simultaneously belong to multiple classes. MLC is increasingly gaining interest in different application domains such as text mining, computer vision, and bioinformatics. Several MLC algorithms have been proposed in the literature, resulting in a meta-optimization problem that the user needs to address: which MLC approach to select for a given dataset? To address this algorithm selection problem, we investigate in this work the quality of an automated approach that uses characteristics of the datasets - so-called features - and a trained algorithm selector to choose which algorithm to apply for a given task. For our empirical evaluation, we use a portfolio of 38 datasets. We consider eight MLC algorithms, whose quality we evaluate using six different performance metrics. We show that our automated algorithm selector outperforms any of the single MLC algorithms, and this is for all evaluated performance measures. Our selection approach is explainable, a characteristic that we exploit to investigate which meta-features have the largest influence on the decisions made by the algorithm selector. Finally, we also quantify the importance of the most significant meta-features for various domains.
Most neural network-based classifiers extract features using several hidden layers and make predictions at the output layer by utilizing these extracted features. We observe that not all features are equally pronounced in all classes; we call such features class-specific features. Existing models do not fully utilize the class-specific differences in features as they feed all extracted features from the hidden layers equally to the output layers. Recent attention mechanisms allow giving different emphasis (or attention) to different features, but these attention models are themselves class-agnostic. In this paper, we propose a novel class-specific attention (CSA) module to capture significant class-specific features and improve the overall classification performance of time series. The CSA module is designed in a way such that it can be adopted in existing neural network (NN) based models to conduct time series classification. In the experiments, this module is plugged into five start-of-the-art neural network models for time series classification to test its effectiveness by using 40 different real datasets. Extensive experiments show that an NN model embedded with the CSA module can improve the base model in most cases and the accuracy improvement can be up to 42%. Our statistical analysis show that the performance of an NN model embedding the CSA module is better than the base NN model on 67% of MTS and 80% of UTS test cases and is significantly better on 11% of MTS and 13% of UTS test cases.
This paper introduces Weighted Optimal Classification Forests (WOCFs), a new family of classifiers that takes advantage of an optimal ensemble of decision trees to derive accurate and interpretable classifiers. We propose a novel mathematical optimization-based methodology which simultaneously constructs a given number of trees, each of them providing a predicted class for the observations in the feature space. The classification rule is derived by assigning to each observation its most frequently predicted class among the trees. We provide a mixed integer linear programming formulation (MIP) for the problem and several novel MIP strengthening / scaling techniques. We report the results of our computational experiments, from which we conclude that our method has equal or superior performance compared with state-of-the-art tree-based classification methods for small to medium-sized instances. We also present three real-world case studies showing that our methodology has very interesting implications in terms of interpretability. Overall, WOCFs complement existing methods such as CART, Optimal Classification Trees, Random Forests and XGBoost. In addition to its Pareto improvement on accuracy and interpretability, we also see unique properties emerging in terms of different trees focusing on different feature variables. This provides nontrivial improvement in interpretability and usability of the trained model in terms of counterfactual explanation. Thus, despite the apparent computational challenge of WOCFs that limit the size of the problems that can be efficiently solved with current MIP, this is an important research direction that can lead to qualitatively different insights for researchers and complement the toolbox of practitioners for high stakes problems.
Point cloud segmentation and classification are some of the primary tasks in 3D computer vision with applications ranging from augmented reality to robotics. However, processing point clouds using deep learning-based algorithms is quite challenging due to the irregular point formats. Voxelization or 3D grid-based representation are different ways of applying deep neural networks to this problem. In this paper, we propose PointResNet, a residual block-based approach. Our model directly processes the 3D points, using a deep neural network for the segmentation and classification tasks. The main components of the architecture are: 1) residual blocks and 2) multi-layered perceptron (MLP). We show that it preserves profound features and structural information, which are useful for segmentation and classification tasks. The experimental evaluations demonstrate that the proposed model produces the best results for segmentation and comparable results for classification in comparison to the conventional baselines.
In this paper, we introduce MINTIME, a video deepfake detection approach that captures spatial and temporal anomalies and handles instances of multiple people in the same video and variations in face sizes. Previous approaches disregard such information either by using simple a-posteriori aggregation schemes, i.e., average or max operation, or using only one identity for the inference, i.e., the largest one. On the contrary, the proposed approach builds on a Spatio-Temporal TimeSformer combined with a Convolutional Neural Network backbone to capture spatio-temporal anomalies from the face sequences of multiple identities depicted in a video. This is achieved through an Identity-aware Attention mechanism that attends to each face sequence independently based on a masking operation and facilitates video-level aggregation. In addition, two novel embeddings are employed: (i) the Temporal Coherent Positional Embedding that encodes each face sequence's temporal information and (ii) the Size Embedding that encodes the size of the faces as a ratio to the video frame size. These extensions allow our system to adapt particularly well in the wild by learning how to aggregate information of multiple identities, which is usually disregarded by other methods in the literature. It achieves state-of-the-art results on the ForgeryNet dataset with an improvement of up to 14% AUC in videos containing multiple people and demonstrates ample generalization capabilities in cross-forgery and cross-dataset settings. The code is publicly available at https://github.com/davide-coccomini/MINTIME-Multi-Identity-size-iNvariant-TIMEsformer-for-Video-Deepfake-Detection.