Deep learning techniques have been used to build velocity models (VMs) for seismic traveltime tomography and have shown encouraging performance in recent years. However, they need to generate labeled samples (i.e., pairs of input and label) to train the deep neural network (NN) with end-to-end learning, and the real labels for field data inversion are usually missing or very expensive. Some traditional tomographic methods can be implemented quickly, but their effectiveness is often limited by prior assumptions. To avoid generating and/or collecting labeled samples, we propose a novel method by integrating deep learning and dictionary learning to enhance the VMs with low resolution by using the traditional tomography-least square method (LSQR). We first design a type of shallow and simple NN to reduce computational cost followed by proposing a two-step strategy to enhance the VMs with low resolution: (1) Warming up. An initial dictionary is trained from the estimation by LSQR through dictionary learning method; (2) Dictionary optimization. The initial dictionary obtained in the warming-up step will be optimized by the NN, and then it will be used to reconstruct high-resolution VMs with the reference slowness and the estimation by LSQR. Furthermore, we design a loss function to minimize traveltime misfit to ensure that NN training is label-free, and the optimized dictionary can be obtained after each epoch of NN training. We demonstrate the effectiveness of the proposed method through the numerical tests on both synthetic and field data.
In inverse problems, it is widely recognized that the incorporation of a sparsity prior yields a regularization effect on the solution. This approach is grounded on the a priori assumption that the unknown can be appropriately represented in a basis with a limited number of significant components, while most coefficients are close to zero. This occurrence is frequently observed in real-world scenarios, such as with piecewise smooth signals. In this study, we propose a probabilistic sparsity prior formulated as a mixture of degenerate Gaussians, capable of modeling sparsity with respect to a generic basis. Under this premise, we design a neural network that can be interpreted as the Bayes estimator for linear inverse problems. Additionally, we put forth both a supervised and an unsupervised training strategy to estimate the parameters of this network. To evaluate the effectiveness of our approach, we conduct a numerical comparison with commonly employed sparsity-promoting regularization techniques, namely LASSO, group LASSO, iterative hard thresholding, and sparse coding/dictionary learning. Notably, our reconstructions consistently exhibit lower mean square error values across all $1$D datasets utilized for the comparisons, even in cases where the datasets significantly deviate from a Gaussian mixture model.
Information Pursuit (IP) is an explainable prediction algorithm that greedily selects a sequence of interpretable queries about the data in order of information gain, updating its posterior at each step based on observed query-answer pairs. The standard paradigm uses hand-crafted dictionaries of potential data queries curated by a domain expert or a large language model after a human prompt. However, in practice, hand-crafted dictionaries are limited by the expertise of the curator and the heuristics of prompt engineering. This paper introduces a novel approach: learning a dictionary of interpretable queries directly from the dataset. Our query dictionary learning problem is formulated as an optimization problem by augmenting IP's variational formulation with learnable dictionary parameters. To formulate learnable and interpretable queries, we leverage the latent space of large vision and language models like CLIP. To solve the optimization problem, we propose a new query dictionary learning algorithm inspired by classical sparse dictionary learning. Our experiments demonstrate that learned dictionaries significantly outperform hand-crafted dictionaries generated with large language models.
Dictionary learning (DL) is commonly used in computational biology to tackle ubiquitous clustering problems due to its conceptual simplicity and relatively low computational complexity. However, DL algorithms produce results that lack interpretability and are not optimized for large-scale graph-structured data. We propose a novel DL algorithm called online convex network dictionary learning (online cvxNDL) that can handle extremely large datasets and enables the interpretation of dictionary elements, which serve as cluster representatives, through convex combinations of real measurements. Moreover, the algorithm can be applied to network-structured data via specialized subnetwork sampling techniques. To demonstrate the utility of our approach, we apply cvxNDL on 3D-genome RNAPII ChIA-Drop data to identify important long-range interaction patterns. ChIA-Drop probes higher-order interactions, and produces hypergraphs whose nodes represent genomic fragments. The hyperedges represent observed physical contacts. Our hypergraph model analysis creates an interpretable dictionary of long-range interaction patterns that accurately represent global chromatin physical contact maps. Using dictionary information, one can also associate the contact maps with RNA transcripts and infer cellular functions. Our results offer two key insights. First, we demonstrate that online cvxNDL retains the accuracy of classical DL methods while simultaneously ensuring unique interpretability and scalability. Second, we identify distinct collections of proximal and distal interaction patterns involving chromatin elements shared by related processes across different chromosomes, as well as patterns unique to specific chromosomes. To associate the dictionary elements with biological properties of the corresponding chromatin regions, we employ Gene Ontology enrichment analysis and perform RNA coexpression studies.
Block majorization-minimization (BMM) is a simple iterative algorithm for nonconvex optimization that sequentially minimizes a majorizing surrogate of the objective function in each block coordinate while the other block coordinates are held fixed. We consider a family of BMM algorithms for minimizing smooth nonconvex objectives, where each parameter block is constrained within a subset of a Riemannian manifold. We establish that this algorithm converges asymptotically to the set of stationary points, and attains an $\epsilon$-stationary point within $\widetilde{O}(\epsilon^{-2})$ iterations. In particular, the assumptions for our complexity results are completely Euclidean when the underlying manifold is a product of Euclidean or Stiefel manifolds, although our analysis makes explicit use of the Riemannian geometry. Our general analysis applies to a wide range of algorithms with Riemannian constraints: Riemannian MM, block projected gradient descent, optimistic likelihood estimation, geodesically constrained subspace tracking, robust PCA, and Riemannian CP-dictionary-learning. We experimentally validate that our algorithm converges faster than standard Euclidean algorithms applied to the Riemannian setting.
Trajectory representation learning on a network enhances our understanding of vehicular traffic patterns and benefits numerous downstream applications. Existing approaches using classic machine learning or deep learning embed trajectories as dense vectors, which lack interpretability and are inefficient to store and analyze in downstream tasks. In this paper, an explainable trajectory representation learning framework through dictionary learning is proposed. Given a collection of trajectories on a network, it extracts a compact dictionary of commonly used subpaths called "pathlets", which optimally reconstruct each trajectory by simple concatenations. The resulting representation is naturally sparse and encodes strong spatial semantics. Theoretical analysis of our proposed algorithm is conducted to provide a probabilistic bound on the estimation error of the optimal dictionary. A hierarchical dictionary learning scheme is also proposed to ensure the algorithm's scalability on large networks, leading to a multi-scale trajectory representation. Our framework is evaluated on two large-scale real-world taxi datasets. Compared to previous work, the dictionary learned by our method is more compact and has better reconstruction rate for new trajectories. We also demonstrate the promising performance of this method in downstream tasks including trip time prediction task and data compression.
Despite their proven success and broad applicability to Electron Microscopy (EM) data, joint dictionary-learning and sparse-coding based inpainting algorithms have so far remained impractical for real-time usage with an Electron Microscope. For many EM applications, the reconstruction time for a single frame is orders of magnitude longer than the data acquisition time, making it impossible to perform exclusively subsampled acquisition. This limitation has led to the development of SenseAI, a C++/CUDA library capable of extremely efficient dictionary-based inpainting. SenseAI provides N-dimensional dictionary learning, live reconstructions, dictionary transfer and visualization, as well as real-time plotting of statistics, parameters, and image quality metrics.
We propose and analyze an efficient algorithm for solving the joint sparse recovery problem using a new regularization-based method, named orthogonally weighted $\ell_{2,1}$ ($\mathit{ow}\ell_{2,1}$), which is specifically designed to take into account the rank of the solution matrix. This method has applications in feature extraction, matrix column selection, and dictionary learning, and it is distinct from commonly used $\ell_{2,1}$ regularization and other existing regularization-based approaches because it can exploit the full rank of the row-sparse solution matrix, a key feature in many applications. We provide a proof of the method's rank-awareness, establish the existence of solutions to the proposed optimization problem, and develop an efficient algorithm for solving it, whose convergence is analyzed. We also present numerical experiments to illustrate the theory and demonstrate the effectiveness of our method on real-life problems.
We present a new algorithm for image segmentation - Level-set KSVD. Level-set KSVD merges the methods of sparse dictionary learning for feature extraction and variational level-set method for image segmentation. Specifically, we use a generalization of the Chan-Vese functional with features learned by KSVD. The motivation for this model is agriculture based. Aerial images are taken in order to detect the spread of fungi in various crops. Our model is tested on such images of cotton fields. The results are compared to other methods.
Many modern machine learning applications - from online principal component analysis to covariance matrix identification and dictionary learning - can be formulated as minimization problems on Riemannian manifolds, and are typically solved with a Riemannian stochastic gradient method (or some variant thereof). However, in many cases of interest, the resulting minimization problem is not geodesically convex, so the convergence of the chosen solver to a desirable solution - i.e., a local minimizer - is by no means guaranteed. In this paper, we study precisely this question, that is, whether stochastic Riemannian optimization algorithms are guaranteed to avoid saddle points with probability 1. For generality, we study a family of retraction-based methods which, in addition to having a potentially much lower per-iteration cost relative to Riemannian gradient descent, include other widely used algorithms, such as natural policy gradient methods and mirror descent in ordinary convex spaces. In this general setting, we show that, under mild assumptions for the ambient manifold and the oracle providing gradient information, the policies under study avoid strict saddle points / submanifolds with probability 1, from any initial condition. This result provides an important sanity check for the use of gradient methods on manifolds as it shows that, almost always, the limit state of a stochastic Riemannian algorithm can only be a local minimizer.
Medical image translation is an ill-posed problem. Unlike existing paired unbounded unidirectional translation networks, in this paper, we consider unpaired medical images and provide a strictly bounded network that yields a stable bidirectional translation. We propose a patch-level concatenated cyclic conditional generative adversarial network (pCCGAN) embedded with adaptive dictionary learning. It consists of two cyclically connected CGANs of 47 layers each; where both generators (each of 32 layers) are conditioned with concatenation of alternate unpaired patches from input and target modality images (not ground truth) of the same organ. The key idea is to exploit cross-neighborhood contextual feature information that bounds the translation space and boosts generalization. The generators are further equipped with adaptive dictionaries learned from the contextual patches to reduce possible degradation. Discriminators are 15-layer deep networks that employ minimax function to validate the translated imagery. A combined loss function is formulated with adversarial, non-adversarial, forward-backward cyclic, and identity losses that further minimize the variance of the proposed learning machine. Qualitative, quantitative, and ablation analysis show superior results on real CT and MRI.
Multiple Instance Learning (MIL) has been widely used in weakly supervised whole slide image (WSI) classification. Typical MIL methods include a feature embedding part, which embeds the instances into features via a pre-trained feature extractor, and an MIL aggregator that combines instance embeddings into predictions. Most efforts have typically focused on improving these parts. This involves refining the feature embeddings through self-supervised pre-training as well as modeling the correlations between instances separately. In this paper, we proposed a sparsely coding MIL (SC-MIL) method that addresses those two aspects at the same time by leveraging sparse dictionary learning. The sparse dictionary learning captures the similarities of instances by expressing them as sparse linear combinations of atoms in an over-complete dictionary. In addition, imposing sparsity improves instance feature embeddings by suppressing irrelevant instances while retaining the most relevant ones. To make the conventional sparse coding algorithm compatible with deep learning, we unrolled it into a sparsely coded module leveraging deep unrolling. The proposed SC module can be incorporated into any existing MIL framework in a plug-and-play manner with an acceptable computational cost. The experimental results on multiple datasets demonstrated that the proposed SC module could substantially boost the performance of state-of-the-art MIL methods. The codes are available at \href{https://github.com/sotiraslab/SCMIL.git}{https://github.com/sotiraslab/SCMIL.git}.
Sketching algorithms have recently proven to be a powerful approach both for designing low-space streaming algorithms as well as fast polynomial time approximation schemes (PTAS). In this work, we develop new techniques to extend the applicability of sketching-based approaches to the sparse dictionary learning and the Euclidean $k$-means clustering problems. In particular, we initiate the study of the challenging setting where the dictionary/clustering assignment for each of the $n$ input points must be output, which has surprisingly received little attention in prior work. On the fast algorithms front, we obtain a new approach for designing PTAS's for the $k$-means clustering problem, which generalizes to the first PTAS for the sparse dictionary learning problem. On the streaming algorithms front, we obtain new upper bounds and lower bounds for dictionary learning and $k$-means clustering. In particular, given a design matrix $\mathbf A\in\mathbb R^{n\times d}$ in a turnstile stream, we show an $\tilde O(nr/\epsilon^2 + dk/\epsilon)$ space upper bound for $r$-sparse dictionary learning of size $k$, an $\tilde O(n/\epsilon^2 + dk/\epsilon)$ space upper bound for $k$-means clustering, as well as an $\tilde O(n)$ space upper bound for $k$-means clustering on random order row insertion streams with a natural "bounded sensitivity" assumption. On the lower bounds side, we obtain a general $\tilde\Omega(n/\epsilon + dk/\epsilon)$ lower bound for $k$-means clustering, as well as an $\tilde\Omega(n/\epsilon^2)$ lower bound for algorithms which can estimate the cost of a single fixed set of candidate centers.
We propose an approach for imaging in scattering media when large and diverse data sets are available. It has two steps. Using a dictionary learning algorithm the first step estimates the true Green's function vectors as columns in an unordered sensing matrix. The array data comes from many sparse sets of sources whose location and strength are not known to us. In the second step, the columns of the estimated sensing matrix are ordered for imaging using Multi-Dimensional Scaling with connectivity information derived from cross-correlations of its columns, as in time reversal. For these two steps to work together we need data from large arrays of receivers so the columns of the sensing matrix are incoherent for the first step, as well as from sub-arrays so that they are coherent enough to obtain the connectivity needed in the second step. Through simulation experiments, we show that the proposed approach is able to provide images in complex media whose resolution is that of a homogeneous medium.
Considering that Coupled Dictionary Learning (CDL) method can obtain a reasonable linear mathematical relationship between resource images, we propose a novel CDL-based Synthetic Aperture Radar (SAR) and multispectral pseudo-color fusion method. Firstly, the traditional Brovey transform is employed as a pre-processing method on the paired SAR and multispectral images. Then, CDL is used to capture the correlation between the pre-processed image pairs based on the dictionaries generated from the source images via enforced joint sparse coding. Afterward, the joint sparse representation in the pair of dictionaries is utilized to construct an image mask via calculating the reconstruction errors, and therefore generate the final fusion image. The experimental verification results of the SAR images from the Sentinel-1 satellite and the multispectral images from the Landsat-8 satellite show that the proposed method can achieve superior visual effects, and excellent quantitative performance in terms of spectral distortion, correlation coefficient, MSE, NIQE, BRISQUE, and PIQE.
This paper presents a method based on a kernel dictionary learning algorithm for segmenting brain tumor regions in magnetic resonance images (MRI). A set of first-order and second-order statistical feature vectors are extracted from patches of size 3 * 3 around pixels in the brain MRI scans. These feature vectors are utilized to train two kernel dictionaries separately for healthy and tumorous tissues. To enhance the efficiency of the dictionaries and reduce training time, a correlation-based sample selection technique is developed to identify the most informative and discriminative subset of feature vectors. This technique aims to improve the performance of the dictionaries by selecting a subset of feature vectors that provide valuable information for the segmentation task. Subsequently, a linear classifier is utilized to distinguish between healthy and unhealthy pixels based on the learned dictionaries. The results demonstrate that the proposed method outperforms other existing methods in terms of segmentation accuracy and significantly reduces both the time and memory required, resulting in a remarkably fast training process.
Dictionary learning is a popular method for obtaining sparse linear representations for high dimensional data, with many applications in image classification, signal processing and machine learning. In this paper, we introduce a novel dictionary learning method based on a recent variable selection algorithm called Feature Selection with Annealing (FSA). Because FSA uses an L0 constraint instead of the L1 penalty, it does not introduce any bias in the coefficients and obtains a more accurate sparse representation. Furthermore, the L0 constraint makes it easy to directly specify the desired sparsity level instead of indirectly through a L1 penalty. Finally, experimental validation on real gray-scale images shows that the proposed method obtains higher accuracy and efficiency in dictionary learning compared to classical methods based on the L1 penalty.
Data-driven approximations of the Koopman operator are promising for predicting the time evolution of systems characterized by complex dynamics. Among these methods, the approach known as extended dynamic mode decomposition with dictionary learning (EDMD-DL) has garnered significant attention. Here we present a modification of EDMD-DL that concurrently determines both the dictionary of observables and the corresponding approximation of the Koopman operator. This innovation leverages automatic differentiation to facilitate gradient descent computations through the pseudoinverse. We also address the performance of several alternative methodologies. We assess a 'pure' Koopman approach, which involves the direct time-integration of a linear, high-dimensional system governing the dynamics within the space of observables. Additionally, we explore a modified approach where the system alternates between spaces of states and observables at each time step -- this approach no longer satisfies the linearity of the true Koopman operator representation. For further comparisons, we also apply a state space approach (neural ODEs). We consider systems encompassing two and three-dimensional ordinary differential equation systems featuring steady, oscillatory, and chaotic attractors, as well as partial differential equations exhibiting increasingly complex and intricate behaviors. Our framework significantly outperforms EDMD-DL. Furthermore, the state space approach offers superior performance compared to the 'pure' Koopman approach where the entire time evolution occurs in the space of observables. When the temporal evolution of the Koopman approach alternates between states and observables at each time step, however, its predictions become comparable to those of the state space approach.
We introduce NPEFF (Non-Negative Per-Example Fisher Factorization), an interpretability method that aims to uncover strategies used by a model to generate its predictions. NPEFF decomposes per-example Fisher matrices using a novel decomposition algorithm that learns a set of components represented by learned rank-1 positive semi-definite matrices. Through a combination of human evaluation and automated analysis, we demonstrate that these NPEFF components correspond to model processing strategies for a variety of language models and text processing tasks. We further show how to construct parameter perturbations from NPEFF components to selectively disrupt a given component's role in the model's processing. Along with conducting extensive ablation studies, we include experiments to show how NPEFF can be used to analyze and mitigate collateral effects of unlearning and use NPEFF to study in-context learning. Furthermore, we demonstrate the advantages of NPEFF over baselines such as gradient clustering and using sparse autoencoders for dictionary learning over model activations.
Transformers are widely used to extract semantic meanings from input tokens, yet they usually operate as black-box models. In this paper, we present a simple yet informative decomposition of hidden states (or embeddings) of trained transformers into interpretable components. For any layer, embedding vectors of input sequence samples are represented by a tensor $\boldsymbol{h} \in \mathbb{R}^{C \times T \times d}$. Given embedding vector $\boldsymbol{h}_{c,t} \in \mathbb{R}^d$ at sequence position $t \le T$ in a sequence (or context) $c \le C$, extracting the mean effects yields the decomposition \[ \boldsymbol{h}_{c,t} = \boldsymbol{\mu} + \mathbf{pos}_t + \mathbf{ctx}_c + \mathbf{resid}_{c,t} \] where $\boldsymbol{\mu}$ is the global mean vector, $\mathbf{pos}_t$ and $\mathbf{ctx}_c$ are the mean vectors across contexts and across positions respectively, and $\mathbf{resid}_{c,t}$ is the residual vector. For popular transformer architectures and diverse text datasets, empirically we find pervasive mathematical structure: (1) $(\mathbf{pos}_t)_{t}$ forms a low-dimensional, continuous, and often spiral shape across layers, (2) $(\mathbf{ctx}_c)_c$ shows clear cluster structure that falls into context topics, and (3) $(\mathbf{pos}_t)_{t}$ and $(\mathbf{ctx}_c)_c$ are mutually nearly orthogonal. We argue that smoothness is pervasive and beneficial to transformers trained on languages, and our decomposition leads to improved model interpretability.