Sequential recommendation aims to predict the next item based on historical interactions. To further enhance the reasoning capability in sequential recommendation, LLMs are employed to predict the next item or generate semantic IDs for item representation, given LLMs' extensive domain knowledge and reasoning ability. However, existing LLM-based methods suffer from two limitations. (i) The scarcity of recommendation data with reasoning paths makes it challenging to design suitable chain-of-thought prompting templates, and the full potential of LLMs' reasoning abilities remains underutilized. (ii) Upon obtaining semantic IDs, the LLMs and their representations are excluded from the subsequent recommendation model training, preventing downstream models from fully utilizing the rich semantic information encoded within these IDs. To address these issues, we propose a novel CoderRec framework, which is capable of fully exploiting the information encoded in semantic IDs to guide the recommendation process. Specifically, to address the problem of scarcity in reasoning path-augmented data, we introduce latent reasoning into sequential recommendation and treat the representation captured by the downstream model as domain-specific latent thought, enabling implicit logical inference without requiring explicit CoT annotations. To ensure that the downstream recommendation models are able to deeply leverage the semantic information within IDs, we propose a novel cross-scale model collaboration strategy, which employs cross-scale IDs and a two-phase approach to align LLM-derived semantics with recommendation objectives. Extensive experiments have shown the effectiveness of our proposed CoderRec framework.

In most domains, anomaly detection is typically cast as an unsupervised learning problem because of the infeasibility of labeling large datasets. In this setup, the evaluation and comparison of different anomaly detection algorithms is difficult. Although some work has been published in this field, they fail to account that different algorithms can detect different kinds of anomalies. More precisely, the literature on this topic has focused on defining criteria to determine which algorithm is better, while ignoring the fact that such criteria are meaningful only if the algorithms being compared are detecting the same kind of anomalies. Therefore, in this article, we propose an equivalence criterion for anomaly detection algorithms that measures to what degree two anomaly detection algorithms detect the same kind of anomalies. First, we lay out a set of desirable properties that such an equivalence criterion should have and why; second, we propose Gaussian Equivalence Criterion (GEC) as equivalence criterion and show mathematically that it has the desirable properties previously mentioned. Finally, we empirically validate these properties using a simulated and a real-world dataset. For the real-world dataset, we show how GEC can provide insight about the anomaly detection algorithms as well as the dataset.

Computed tomography (CT) is one of the most important medical imaging technologies in use today. Most commercial CT products use a technique known as the filtered backprojection (FBP) that is fast and can produce decent image quality when an X-ray dose is high. However, the FBP is not good enough on low-dose X-ray CT imaging because the CT image reconstruction problem becomes more stochastic. A more effective reconstruction technique proposed recently and implemented in a limited number of CT commercial products is an iterative reconstruction (IR). The IR technique is based on a Bayesian formulation of the CT image reconstruction problem with an explicit model of the CT scanning, including its stochastic nature, and a prior model that incorporates our knowledge about what a good CT image should look like. However, constructing such prior knowledge is more complicated than it seems. In this article, we propose a novel neural network for CT image reconstruction. The network is based on the IR formulation and constructed with a recurrent neural network (RNN). Specifically, we transform the gated recurrent unit (GRU) into a neural network performing CT image reconstruction. We call it "GRU reconstruction." This neural network conducts concurrent dual-domain learning. Many deep learning (DL)-based methods in medical imaging are single-domain learning, but dual-domain learning performs better because it learns from both the sinogram and the image domain. In addition, we propose backpropagation through stage (BPTS) as a new RNN backpropagation algorithm. It is similar to the backpropagation through time (BPTT) of an RNN; however, it is tailored for iterative optimization. Results from extensive experiments indicate that our proposed method outperforms conventional model-based methods, single-domain DL methods, and state-of-the-art DL techniques in terms of the root mean squared error (RMSE), the peak signal-to-noise ratio (PSNR), and the structure similarity (SSIM) and in terms of visual appearance.

As we reach the end of Moore’s Law and Dennard Scaling, it has become highly desirable to design a highly integrated and optimized pipeline specifically for computer vision. A new generation of integrated "smart" visual processors that streamline an end-to-end optimized visual information acquisition and processing pipeline (VIAPP) becomes necessary to lower the cost, power consumption, and latency.We describe a new paradigm for VIAPP as Vision Perception Unit (VPU), wherein electric signals generated by photons are amplified before converting to the digital signals to emulate an initial layer of a convolutional neural network (CNN). The outputs from these layers are then converted to digital signals and processed by following layers of a deep CNN.

Tracking-by-detection is one of the most popular approaches to tracking multiple objects in which the detector plays an important role. Sometimes, detector failures caused by occlusions or various poses are unavoidable and lead to tracking failure. To cope with this problem, we construct a heterogeneous association graph that fuses high-level detections and low-level image evidence for target association. Compared with other methods using low-level information, our proposed heterogeneous association fusion (HAF) tracker is less sensitive to particular parameters and is easier to extend and implement. We use the fused association graph to build track trees for HAF and solve them by the multiple hypotheses tracking framework, which has been proven to be competitive by introducing efficient pruning strategies. In addition, the novel idea of adaptive weights is proposed to analyze the contribution between motion and appearance. We also evaluated our results on the MOT challenge benchmarks and achieved state-of-the-art results on the MOT Challenge 2017.

Since a light-field camera is able to capture more information than a traditional camera, a lot of methods, such as depth estimation, image super-resolution, and view synthesis, are explored for recovering scene information. In this paper, we propose a novel framework for scene recovery based on lenslet-based light-field camera images. Instead of using traditional matching terms, we design a new micro-lens-based matching term to calculate structure information and recover several kinds of scene information simultaneously. On the one hand, inherent information in micro-lens images is selected to complement details in sub-aperture images. On the other hand, sub-aperture images are used to expand micro-lens images and synthesize new view images. A new micro-lens-based consistency metric is introduced for the matching term to handle occlusions in depth estimation and image reconstruction. The newly appeared and newly occluded areas in synthesized views are analyzed and recovered based on information from surrounding points. Experimental results show that the proposed depth estimation method outperforms state-of-the-art methods on both synthetic and lenslet-based light-field images, especially in low-texture and occlusion regions. Furthermore, the super-resolution and view synthesis methods are able to acquire view images with more details and less aliasing artifacts.

The Transport Control Protocol (TCP) has been widely used by wired and wireless Internet applications such as FTP, email and HTTP. Numerous congestion algorithms have been proposed to improve the performance of TCP in various scenarios, especially for high bandwidth-delay product (BDP) and wireless networks. Although different algorithms may achieve different performance improvements under different network conditions, designing a congestion algorithm that performs well across a wide spectrum of network conditions remains a great challenge. In this paper, we propose a novel congestion control algorithm, named TCP-FIT, which could perform gracefully in both wireless and high BDP networks. The algorithm was inspired by parallel TCP, but with the important distinctions that only one TCP connection with one congestion window is established for each TCP session, and that no modifications to other layers (e.g. the application layer) of the end-to-end system need to be made. Extensive experimental results obtained using both network simulators as well as over “live” wired line, WiFi and 3G networks at different geographical locations and at different times of the day are presented. The performance of the algorithm shown in the experiment results is significantly improved as compared to other state-of-the-art algorithms, while maintaining good fairness.

This paper considers constrained lscr1minimization methods in a unified framework for the recovery of high-dimensional sparse signals in three settings: noiseless, bounded error, and Gaussian noise. Both lscr1minimization with an lscrinfinconstraint (Dantzig selector) and lscr1minimization under anllscr2constraint are considered. The results of this paper improve the existing results in the literature by weakening the conditions and tightening the error bounds. The improvement on the conditions shows that signals with larger support can be recovered accurately. In particular, our results illustrate the relationship between lscr1minimization with anllscr2constraint and lscr1minimization with an lscrinfinconstraint. This paper also establishes connections between restricted isometry property and the mutual incoherence property. Some results of Candes, Romberg, and Tao (2006), Candes and Tao (2007), and Donoho, Elad, and Temlyakov (2006) are extended.

Prediction is an essential operation in many image processing applications, such as object detection and image and video compression. When the images are modeled as Gaussian, the optimal predictor is linear and easy to obtain. However, image texture and clutter are often non-Gaussian, and, in such cases, optimal predictors are difficult to obtain. In this paper, we derive an optimal predictor for an important class of non-Gaussian image models, the block-based multivariate Gaussian mixture model. This predictor has a special nonlinear structure: it is a linear combination of the neighboring pixels, but the combination coefficients are also functions of the neighboring pixels, not constants. The efficacy of this predictor is demonstrated in object detection experiments where the prediction error image is used to identify "hidden" objects. Experimental results indicate that when the background texture is nonlinear, i.e., with fast-switching gray-level patches, it performs significantly better than the optimal linear predictor.

With the rising popularity of network-based applications and the potential use of mobile ad hoc networks in civilian life, an efficient resource discovery service is needed in such networks for quickly locating resource providers. In addition, to improve user experience, QoS awareness is also crucial. We identify the challenges when basic resource discovery techniques for the Internet are used in mobile ad hoc networks. We then propose a framework that provides a unified solution to the discovery of resources and QoS-aware selection of resource providers. The key entities of this framework are a set of self-organized discovery agents. These agents manage the directory information of resources using hash indexing. They also dynamically partition the network into domains and collect intra- and inter-domain QoS information to select appropriate providers. Simulation results show that our framework improves the QoS delivered to the clients, while the cost and response time are kept at a low level.

We describe a novel approach for image sequence segmentation. It contains three parts: global motion compensation, robust frame differencing, and curve evolution. It is computationally efficient, does not require dense-field motion estimation, and is insensitive to noise and global/background motion. It works for black-and-white and color image sequences. The efficacy of this approach is demonstrated on both TV and surveillance image sequences.

Describes a novel technique for color image segmentation. This technique includes three components: a color space transformation; MRF-EM segmentation; and a multiresolution implementation termed "narrowband". The color space transformation, from RGB to LUV, contributes to a perceptually reasonable segmentation result. The MRF-EM algorithm permits unsupervised segmentation and enforces a region smoothness constraint. The narrow-band, multiresolution implementation confines fine resolution processing to a small fraction of the image, thereby achieving significant acceleration over traditional multiresolution schemes. Experimental results on dermatoscopic images demonstrate the efficacy of the proposed technique.

This correspondence describes a novel nonlinear scheme for figure-ground separation where a nonlinear differential equation is defined at each pixel site and coupled with those at neighboring sites. The steady state solution enhances salient image structures and suppresses background noise. Experimental results on synthetic and real-world images demonstrate the efficacy of this scheme.

In this correspondence, a solution is developed for the regularized total least squares (RTLS) estimate in linear inverse problems where the linear operator is nonconvolutional. Our approach is based on a Rayleigh quotient (RQ) formulation of the TLS problem, and we accomplish regularization by modifying the RQ function to enforce a smooth solution. A conjugate gradient algorithm is used to minimize the modified RQ function. As an example, the proposed approach has been applied to the perturbation equation encountered in optical tomography. Simulation results show that this method provides more stable and accurate solutions than the regularized least squares and a previously reported total least squares approach, also based on the RQ formulation.

Image segmentation is an important and early processing stage in many image analysis problems. Often, this must be done in an unsupervised fashion in that training data is not available and the class-conditioned feature vectors must be estimated directly from the data. A major problem in such applications is the determination of the number of classes actually present in an image. This problem, called the cluster validation problem, remains essentially unsolved. We investigate the cluster validation problem associated with the use of a previously developed unsupervised segmentation algorithm based upon the expectation-maximization (EM) algorithm. More specifically, we consider several well-known information-theoretic criteria (ICs) as candidate solutions to the validation problem when used in conjunction with this EM-based segmentation scheme. We show that these criteria generally provide inappropriate solutions due to the domination of the penalty term by the associated log-likelihood function. As an alternative we propose a model-fitting technique in which the complete data log-likelihood functional is modeled as an exponential function in the number of classes acting. The estimated number of classes are then determined in a manner similar to finding the rise time of the exponential function. This new validation technique is shown to be robust and outperform the ICs in our experiments. Experimental results for both synthetic and real world imagery are detailed.

A multiresolution statistical model, consisting of random fields in wavelet subbands, is proposed for texture, and has produced promising results in texture synthesis experiments. The basic idea here is that a complex random field model, e.g., one that contains long-range and nonlinear spatial correlations, can be constructed from several simpler ones.

A regularized total least square (RTLS) approach to solve a linear perturbation equation encountered in optical tomography is developed based on the Rayleigh quotient formulation. To compute efficiently the solution, the Rayleigh quotient form of the RTLS filter (RQF-RTLS) is used and a conjugate gradient algorithm is implemented. Simulation results show that the RQF-RTLS method obtains more stable and accurate solutions than the regularized least squares (RLS) approach which does not account for the errors in the operator.

The Gibbs-Bogoliubov-Feynman (GBF) inequality of statistical mechanics is adopted, with an information-theoretic interpretation, as a general optimization framework for deriving and examining various mean field approximations for Markov random fields (MRF's). The efficacy of this approach is demonstrated through the compound Gauss-Markov (CGM) model, comparisons between different mean field approximations, and experimental results in image restoration.

Convergence of the mean field equations is analyzed for the Ising model with a nonhomogeneous external field. When beta is small, the equations are a contraction mapping and thereby convergent. When beta is large, however, they can be noncontracting and in this case, contraction is restored by a slight modification of the equations.

This paper describes a neural network inspired dynamical system approach to a perceptual grouping problem-figure-ground separation. In this approach, a non-linear differential equation is defined at each pixel site and coupled with those at its neighbours. The steady state solution would determine whether a pixel is part of a salient structure or background/noise. The neighbourhood couplings are used to achieve spatial interactions that are essential to perceptual grouping, such as excitation and inhibition. Experimental results on the grouping of dots in synthetic and real-world images demonstrate the efficacy of the proposed approach.

The efficacy of the compound Gauss-Markov (CGM) model, initially proposed by Jeng and Woods (1990), has been demonstrated in several image processing applications. However, parameter estimation for the CGM model is difficult since it is not clear as to how the constraints or interdependence amongst the model parameters can be incorporated into the estimation procedures. As result, the parameter estimates tend to be inconsistent. It is shown that, under some reasonable symmetry constraints, the 80 interdependent parameters of the CGM model can be reduced to seven independent ones. This guarantees the consistency of model parameters obtained from parameter estimation algorithms, thereby removing a main obstacle for the parameter estimation of the CGM model.

Previously, Markov random field (MRF) model-based techniques have been proposed for image motion estimation. Since motion estimation is usually an ill-posed problem, various constraints are needed to obtain a unique and stable solution. The main advantage of the MRF approach is its capacity to incorporate such constraints, for instance, motion continuity within an object and motion discontinuity at the boundaries between objects. In the MRF approach, motion estimation is often formulated as an optimization problem, and two frequently used optimization methods are simulated annealing (SA) and iterative-conditional mode (ICM). Although the SA is theoretically optimal in the sense of finding the global optimum, it usually takes many iterations to converge. The ICM, on the other hand, converges quickly, but its results are often unsatisfactory due to its "hard decision" nature. Previously, the authors have applied the mean field theory to image segmentation and image restoration problems. It provides results nearly as good as SA but with much faster convergence. The present paper shows how the mean field theory can be applied to MRF model-based motion estimation. This approach is demonstrated on both synthetic and real-world images, where it produced good motion estimates.

Image segmentation is an important processing stage in many image analysis problems. Often this must be done in an unsupervised fashion in that training data is not available. A major obstacle in such applications is the determination of the number of distinct regions present in an image. This problem, called the cluster validation problem, remains essentially unsolved. We investigate the cluster validation problem associated with the use of a previously developed unsupervised segmentation algorithm based upon the expectation-maximization (EM) algorithm. We consider several well-known information-theoretic criteria (ICs) as candidate solutions. We show that these criteria generally provide inappropriate results. As an alternative we propose a model-fitting technique in which the complete data log-likelihood functional is modeled as an exponential function in the number of classes acting, and the class estimate is related to the rise time. This new validation technique is shown to be robust and outperform the ICs in our experiments.>

An unsupervised stochastic model-based approach to image segmentation is described, and some of its properties investigated. In this approach, the problem of model parameter estimation is formulated as a problem of parameter estimation from incomplete data, and the expectation-maximization (EM) algorithm is used to determine a maximum-likelihood (ML) estimate. Previously, the use of the EM algorithm in this application has encountered difficulties since an analytical expression for the conditional expectations required in the EM procedure is generally unavailable, except for the simplest models. In this paper, two solutions are proposed to solve this problem: a Monte Carlo scheme and a scheme related to Besag's (1986) iterated conditional mode (ICM) method. Both schemes make use of Markov random-field modeling assumptions. Examples are provided to illustrate the implementation of the EM algorithm for several general classes of image models. Experimental results on both synthetic and real images are provided.

A Markov random field (MRF) model-based EM (expectation-maximization) procedure for simultaneously estimating the degradation model and restoring the image is described. The MRF is a coupled one which provides continuity (inside regions of smooth gray tones) and discontinuity (at region boundaries) constraints for the restoration problem which is, in general, ill posed. The computational difficulty associated with the EM procedure for MRFs is resolved by using the mean field theory from statistical mechanics. An orthonormal blur decomposition is used to reduce the chances of undesirable locally optimal estimates. Experimental results on synthetic and real-world images show that this approach provides good blur estimates and restored images. The restored images are comparable to those obtained by a Wiener filter in mean-square error, but are most visually pleasing.

The application of a Markov random field (MRF) state model in an expectation-maximization (EM)-based approach to unsupervised image segmentation is investigated. In the calculation of the marginal distribution of the state field, it is shown that the use of the expected state values for interacting pixel sites in the computation of the MRF energy function may be interpreted as a mean-field approximation. The implications of calculating a self-consistent expectation of the state field are considered. EM convergence criteria are considered, and a criterion based upon divergence is proposed. Experimental results based on synthetic data illustrate the performance advantage of the mean-field approximation and the computational advantage of using self-consistent expectations.>

An image is segmented into a collection of disjoint regions that form the nodes of an adjacency graph, and image interpretation is achieved through assigning object labels (or interpretations) to the segmented regions (or nodes) using domain knowledge, extracted feature measurements, and spatial relationships between the various regions. The interpretation labels are modeled as a Markov random field (MRF) on the corresponding adjacency graph, and the image interpretation problem is then formulated as a maximum a posteriori (MAP) estimation rule, given domain knowledge and region-based measurements. Simulated annealing is used to find this best realization or optimal MAP interpretation. This approach also provides a systematic method for organizing and representing domain knowledge through appropriate design of the clique functions describing the Gibbs distribution representing the pdf of the underlying MRF. A general methodology is provided for the design of the clique functions. Results of image interpretation experiments on synthetic and real-world images are described.>

A clustering scheme is used for model parameter estimation. Most of the existing clustering procedures require prior knowledge of the number of classes, which is often, as in unsupervised image segmentation, unavailable and must be estimated. This problem is known as the cluster validation problem. For unsupervised image segmentation the solution of this problem directly affects the quality of the segmentation. A model-fitting approach to the cluster validation problem based on Akaike's information criterion is proposed, and its efficacy and robustness are demonstrated through experimental results for synthetic mixture data and image data.>

A Markov random field (MRF) model-based approach to automated image interpretation is described and demonstrated as a region-based scheme. In this approach, an image is first segmented into a collection of disjoint regions which form the nodes of an adjacency graph. Image interpretation is then achieved through assigning object labels, or interpretations, to the segmented regions, or nodes, using domain knowledge, extracted feature measurements, and spatial relationships between the various regions. The interpretation labels are modeled as a MRF on the corresponding adjacency graph, and the image interpretation problem are formulated as a maximum a posteriori estimation rule. Simulated annealing is used to find the best realization, or optimal interpretation. Through the MRF model, this approach also provides a systematic method for organizing and representing domain knowledge through the clique functions of the probability density function underlying MRF. Results of image interpretation experiments performed on synthetic and real-world images using this approach are described.>

A model-fitting approach to the cluster validation problem based upon Akaike's information criterion (AIC) is proposed. The explicit evaluation of the AIC for the image segmentation problem is achieved through an approximate maximum-likelihood-estimation algorithm. The efficacy of the proposed approach is demonstrated through experimental results for both synthetic mixture data, where the number of clusters is known, and stochastic model-based image segmentation operating on real-world images, for which the number of clusters is unknown. This approach is shown to correctly identify the known number of clusters in the synthetically generated data and to result in good subjective segmentations in aerial photographs.>