Search

Distributed word representation (a.k.a. word embedding) is a key focus in natural language processing (NLP). As a highly successful word embedding model, Word2Vec offers an efficient method for learning distributed word representations on large datasets. However, Word2Vec lacks consideration for distances between center and context words. We propose two novel methods, Learnable Formulated Weights (LFW) and Epoch-based Dynamic Window Size (EDWS), to incorporate distance information into two variants of Word2Vec, the Continuous Bag-of-Words (CBOW) model and the Continuous Skip-gram (Skip-gram) model. For CBOW, LFW uses a formula with learnable parameters that best reflects the relationship of influence and distance between words to calculate distance-related weights for average pooling, providing insights for future NLP text modeling research. For Skip-gram, we improve its dynamic window size strategy to introduce distance information in a more balanced way. Experiments prove the effectiveness of LFW and EDWS in enhancing Word2Vec's performance, surpassing previous state-of-the-art methods.

Sparse Partial Least Squares (sPLS) is a common dimensionality reduction technique for data fusion, which projects data samples from two views by seeking linear combinations with a small number of variables with the maximum variance. However, sPLS extracts the combinations between two data sets with all data samples so that it cannot detect latent subsets of samples. To extend the application of sPLS by identifying a specific subset of samples and remove outliers, we propose an $\ell_\infty/\ell_0$-norm constrained weighted sparse PLS ($\ell_\infty/\ell_0$-wsPLS) method for joint sample and feature selection, where the $\ell_\infty/\ell_0$-norm constrains are used to select a subset of samples. We prove that the $\ell_\infty/\ell_0$-norm constrains have the Kurdyka-\L{ojasiewicz}~property so that a globally convergent algorithm is developed to solve it. Moreover, multi-view data with a same set of samples can be available in various real problems. To this end, we extend the $\ell_\infty/\ell_0$-wsPLS model and propose two multi-view wsPLS models for multi-view data fusion. We develop an efficient iterative algorithm for each multi-view wsPLS model and show its convergence property. As well as numerical and biomedical data experiments demonstrate the efficiency of the proposed methods.

This paper introduces a novel explainable image quality evaluation approach called X-IQE, which leverages visual large language models (LLMs) to evaluate text-to-image generation methods by generating textual explanations. X-IQE utilizes a hierarchical Chain of Thought (CoT) to enable MiniGPT-4 to produce self-consistent, unbiased texts that are highly correlated with human evaluation. It offers several advantages, including the ability to distinguish between real and generated images, evaluate text-image alignment, and assess image aesthetics without requiring model training or fine-tuning. X-IQE is more cost-effective and efficient compared to human evaluation, while significantly enhancing the transparency and explainability of deep image quality evaluation models. We validate the effectiveness of our method as a benchmark using images generated by prevalent diffusion models. X-IQE demonstrates similar performance to state-of-the-art (SOTA) evaluation methods on COCO Caption, while overcoming the limitations of previous evaluation models on DrawBench, particularly in handling ambiguous generation prompts and text recognition in generated images. Project website: https://github.com/Schuture/Benchmarking-Awesome-Diffusion-Models, Comment: 18 pages, 6 tables, 11 figures, NeurIPS 2023 submission

Well-annotated medical datasets enable deep neural networks (DNNs) to gain strong power in extracting lesion-related features. Building such large and well-designed medical datasets is costly due to the need for high-level expertise. Model pre-training based on ImageNet is a common practice to gain better generalization when the data amount is limited. However, it suffers from the domain gap between natural and medical images. In this work, we pre-train DNNs on ultrasound (US) domains instead of ImageNet to reduce the domain gap in medical US applications. To learn US image representations based on unlabeled US videos, we propose a novel meta-learning-based contrastive learning method, namely Meta Ultrasound Contrastive Learning (Meta-USCL). To tackle the key challenge of obtaining semantically consistent sample pairs for contrastive learning, we present a positive pair generation module along with an automatic sample weighting module based on meta-learning. Experimental results on multiple computer-aided diagnosis (CAD) problems, including pneumonia detection, breast cancer classification, and breast tumor segmentation, show that the proposed self-supervised method reaches state-of-the-art (SOTA). The codes are available at https://github.com/Schuture/Meta-USCL.

Deep transfer learning (DTL) has formed a long-term quest toward enabling deep neural networks (DNNs) to reuse historical experiences as efficiently as humans. This ability is named knowledge transferability. A commonly used paradigm for DTL is firstly learning general knowledge (pre-training) and then reusing (fine-tuning) them for a specific target task. There are two consensuses of transferability of pre-trained DNNs: (1) a larger domain gap between pre-training and downstream data brings lower transferability; (2) the transferability gradually decreases from lower layers (near input) to higher layers (near output). However, these consensuses were basically drawn from the experiments based on natural images, which limits their scope of application. This work aims to study and complement them from a broader perspective by proposing a method to measure the transferability of pre-trained DNN parameters. Our experiments on twelve diverse image classification datasets get similar conclusions to the previous consensuses. More importantly, two new findings are presented, i.e., (1) in addition to the domain gap, a larger data amount and huge dataset diversity of downstream target task also prohibit the transferability; (2) although the lower layers learn basic image features, they are usually not the most transferable layers due to their domain sensitivity.

The permutation flow shop scheduling (PFSS), aiming at finding the optimal permutation of jobs, is widely used in manufacturing systems. When solving large-scale PFSS problems, traditional optimization algorithms such as heuristics could hardly meet the demands of both solution accuracy and computational efficiency, thus learning-based methods have recently garnered more attention. Some work attempts to solve the problems by reinforcement learning methods, which suffer from slow convergence issues during training and are still not accurate enough regarding the solutions. To that end, we propose to train the model via expert-driven imitation learning, which accelerates convergence more stably and accurately. Moreover, in order to extract better feature representations of input jobs, we incorporate the graph structure as the encoder. The extensive experiments reveal that our proposed model obtains significant promotion and presents excellent generalizability in large-scale problems with up to 1000 jobs. Compared to the state-of-the-art reinforcement learning method, our model's network parameters are reduced to only 37\% of theirs, and the solution gap of our model towards the expert solutions decreases from 6.8\% to 1.3\% on average. The code is available at: \url{https://github.com/longkangli/PFSS-IL}., Comment: 6 figures, 11 tables, AAAI 2024

In medical image analysis, transfer learning is a powerful method for deep neural networks (DNNs) to generalize well on limited medical data. Prior efforts have focused on developing pre-training algorithms on domains such as lung ultrasound, chest X-ray, and liver CT to bridge domain gaps. However, we find that model fine-tuning also plays a crucial role in adapting medical knowledge to target tasks. The common fine-tuning method is manually picking transferable layers (e.g., the last few layers) to update, which is labor-expensive. In this work, we propose a meta-learning-based LR tuner, named MetaLR, to make different layers automatically co-adapt to downstream tasks based on their transferabilities across domains. MetaLR learns appropriate LRs for different layers in an online manner, preventing highly transferable layers from forgetting their medical representation abilities and driving less transferable layers to adapt actively to new domains. Extensive experiments on various medical applications show that MetaLR outperforms previous state-of-the-art (SOTA) fine-tuning strategies. Codes are released., Comment: MICCAI 2023

A major challenge in Fine-Grained Visual Classification (FGVC) is distinguishing various categories with high inter-class similarity by learning the feature that differentiate the details. Conventional cross entropy trained Convolutional Neural Network (CNN) fails this challenge as it may suffer from producing inter-class invariant features in FGVC. In this work, we innovatively propose to regularize the training of CNN by enforcing the uniqueness of the features to each category from an information theoretic perspective. To achieve this goal, we formulate a minimax loss based on a game theoretic framework, where a Nash equilibria is proved to be consistent with this regularization objective. Besides, to prevent from a feasible solution of minimax loss that may produce redundant features, we present a Feature Redundancy Loss (FRL) based on normalized inner product between each selected feature map pair to complement the proposed minimax loss. Superior experimental results on several influential benchmarks along with visualization show that our method gives full play to the performance of the baseline model without additional computation and achieves comparable results with state-of-the-art models.

Singular value decomposition (SVD) is the mathematical basis of principal component analysis (PCA). Together, SVD and PCA are one of the most widely used mathematical formalism/decomposition in machine learning, data mining, pattern recognition, artificial intelligence, computer vision, signal processing, etc. In recent applications, regularization becomes an increasing trend. In this paper, we present a regularized SVD (RSVD), present an efficient computational algorithm, and provide several theoretical analysis. We show that although RSVD is non-convex, it has a closed-form global optimal solution. Finally, we apply RSVD to the application of recommender system and experimental result show that RSVD outperforms SVD significantly.

Support Vector Machine (SVM) is an efficient classification approach, which finds a hyperplane to separate data from different classes. This hyperplane is determined by support vectors. In existing SVM formulations, the objective function uses L2 norm or L1 norm on slack variables. The number of support vectors is a measure of generalization errors. In this work, we propose a Minimal SVM, which uses L0.5 norm on slack variables. The result model further reduces the number of support vectors and increases the classification performance.

In many real-world applications, data usually contain outliers. One popular approach is to use L2,1 norm function as a robust error/loss function. However, the robustness of L2,1 norm function is not well understood so far. In this paper, we propose a new Vector Outlier Regularization (VOR) framework to understand and analyze the robustness of L2,1 norm function. Our VOR function defines a data point to be outlier if it is outside a threshold with respect to a theoretical prediction, and regularize it-pull it back to the threshold line. We then prove that L2,1 function is the limiting case of this VOR with the usual least square/L2 error function as the threshold shrinks to zero. One interesting property of VOR is that how far an outlier lies away from its theoretically predicted value does not affect the final regularization and analysis results. This VOR property unmasks one of the most peculiar property of L2,1 norm function: The effects of outliers seem to be independent of how outlying they are-if an outlier is moved further away from the intrinsic manifold/subspace, the final analysis results do not change. VOR provides a new way to understand and analyze the robustness of L2,1 norm function. Applying VOR to matrix factorization leads to a new VORPCA model. We give a comprehensive comparison with trace-norm based L21-norm PCA to demonstrate the advantages of VORPCA.

In many real-world applications, data come with corruptions, large errors or outliers. One popular approach is to use L1-norm function. However, the robustness of L1-norm function is not well understood so far. In this paper, we present a new outlier regularization framework to understand and analyze the robustness of L1-norm function. There are two main features for the proposed outlier regularization. (1) A key property of outlier regularization is that how far an outlier lies away from its theoretically predicted value does not affect the final regularization and analysis results. (2) Another important feature of outlier regularization is that it has an equivalent continuous representation that closely relates to L1 function. This provides a new way to understand and analyze the robustness of L1 function. We apply our outlier regularization framework to PCA and propose an outlier regularized PCA (ORPCA) model. Comparing to the trace-normbased robust PCA, ORPCA has several benefits: (1) It does not suffer singular value suppression. (2) It can retain small high rank components which help retain fine details of data. (3) ORPCA can be computed more efficiently.

In many real-world applications, image data often come with noises, corruptions or large errors. One approach to deal with noise image data is to use data recovery techniques which aim to recover the true uncorrupted signals from the observed noise images. In this paper, we first introduce a novel corruption recovery transformation (CRT) model which aims to recover multiple (or a collection of) corrupted images using a single affine transformation. Then, we show that the introduced CRT can be efficiently constructed through learning from training data. Once CRT is learned, we can recover the true signals from the new incoming/test corrupted images explicitly. As an application, we apply our CRT to image recognition task. Experimental results on six image datasets demonstrate that the proposed CRT model is effective in recovering noise image data and thus leads to better recognition results.

Real life data often includes information from different channels. For example, in computer vision, we can describe an image using different image features, such as pixel intensity, color, HOG, GIST feature, SIFT features, etc.. These different aspects of the same objects are often called multi-view (or multi-modal) data. Low-rank regression model has been proved to be an effective learning mechanism by exploring the low-rank structure of real life data. But previous low-rank regression model only works on single view data. In this paper, we propose a multi-view low-rank regression model by imposing low-rank constraints on multi-view regression model. Most importantly, we provide a closed-form solution to the multi-view low-rank regression model. Extensive experiments on 4 multi-view datasets show that the multi-view low-rank regression model outperforms single-view regression model and reveals that multi-view low-rank structure is very helpful., Comment: Twenty-Ninth AAAI Conference on Artificial Intelligence, AAAI, 2015

Kernel alignment measures the degree of similarity between two kernels. In this paper, inspired from kernel alignment, we propose a new Linear Discriminant Analysis (LDA) formulation, kernel alignment LDA (kaLDA). We first define two kernels, data kernel and class indicator kernel. The problem is to find a subspace to maximize the alignment between subspace-transformed data kernel and class indicator kernel. Surprisingly, the kernel alignment induced kaLDA objective function is very similar to classical LDA and can be expressed using between-class and total scatter matrices. This can be extended to multi-label data. We use a Stiefel-manifold gradient descent algorithm to solve this problem. We perform experiments on 8 single-label and 6 multi-label data sets. Results show that kaLDA has very good performance on many single-label and multi-label problems., Comment: Joint European Conference on Machine Learning and Knowledge Discovery in Databases, ECML PKDD, 2014

Linear Discriminant Analysis (LDA) is a widely-used supervised dimensionality reduction method in computer vision and pattern recognition. In null space based LDA (NLDA), a well-known LDA extension, between-class distance is maximized in the null space of the within-class scatter matrix. However, there are some limitations in NLDA. Firstly, for many data sets, null space of within-class scatter matrix does not exist, thus NLDA is not applicable to those datasets. Secondly, NLDA uses arithmetic mean of between-class distances and gives equal consideration to all between-class distances, which makes larger between-class distances can dominate the result and thus limits the performance of NLDA. In this paper, we propose a harmonic mean based Linear Discriminant Analysis, Multi-Class Discriminant Analysis (MCDA), for image classification, which minimizes the reciprocal of weighted harmonic mean of pairwise between-class distance. More importantly, MCDA gives higher priority to maximize small between-class distances. MCDA can be extended to multi-label dimension reduction. Results on 7 single-label data sets and 4 multi-label data sets show that MCDA has consistently better performance than 10 other single-label approaches and 4 other multi-label approaches in terms of classification accuracy, macro and micro average F1 score., Comment: IEEE 28th International Conference on Tools with Artificial Intelligence, ICTAI 2016

Deep Learning is a very powerful machine learning model. Deep Learning trains a large number of parameters for multiple layers and is very slow when data is in large scale and the architecture size is large. Inspired from the shrinking technique used in accelerating computation of Support Vector Machines (SVM) algorithm and screening technique used in LASSO, we propose a shrinking Deep Learning with recall (sDLr) approach to speed up deep learning computation. We experiment shrinking Deep Learning with recall (sDLr) using Deep Neural Network (DNN), Deep Belief Network (DBN) and Convolution Neural Network (CNN) on 4 data sets. Results show that the speedup using shrinking Deep Learning with recall (sDLr) can reach more than 2.0 while still giving competitive classification performance., Comment: The 22nd IEEE International Conference on Parallel and Distributed Systems (ICPADS 2016)

Local Linear embedding (LLE) is a popular dimension reduction method. In this paper, we first show LLE with nonnegative constraint is equivalent to the widely used Laplacian embedding. We further propose to iterate the two steps in LLE repeatedly to improve the results. Thirdly, we relax the kNN constraint of LLE and present a sparse similarity learning algorithm. The final Iterative LLE combines these three improvements. Extensive experiment results show that iterative LLE algorithm significantly improve both classification and clustering results., Comment: Appears in Proceedings of the 29th International Conference on Machine Learning (ICML 2012)

For tensor decompositions such as HOSVD and ParaFac, the objective functions are nonconvex. This implies, theoretically, there exists a large number of local optimas: starting from different starting point, the iteratively improved solution will converge to different local solutions. This non-uniqueness present a stability and reliability problem for image compression and retrieval. In this paper, we present the results of a comprehensive investigation of this problem. We found that although all tensor decomposition algorithms fail to reach a unique global solution on random data and severely scrambled data; surprisingly however, on all real life several data sets (even with substantial scramble and occlusions), HOSVD always produce the unique global solution in the parameter region suitable to practical applications, while ParaFac produce non-unique solutions. We provide an eigenvalue based rule for the assessing the solution uniqueness., Comment: Submitted to CVPR2009 in Nov. 20, 2008