Adapting large language models (LLMs) to specific domains often faces a critical bottleneck: the scarcity of high-quality, human-curated data. While large volumes of unchecked data are readily available, indiscriminately using them for fine-tuning risks introducing noise and degrading performance. Strategic data selection is thus crucial, requiring a method that is both accurate and efficient. Existing approaches, categorized as similarity-based and direct optimization methods, struggle to simultaneously achieve these goals. In this paper, we introduce LAMDAS (LLM as an implicit classifier for domain-specific Data Selection), a novel approach that leverages the pre-trained LLM itself as an implicit classifier, thereby bypassing explicit feature engineering and computationally intensive optimization process. LAMDAS reframes data selection as a one-class classification problem, identifying candidate data that "belongs" to the target domain defined by a small reference dataset. Extensive experimental results demonstrate that LAMDAS not only exceeds the performance of full-data training using a fraction of the data but also outperforms nine state-of-the-art (SOTA) baselines under various scenarios. Furthermore, LAMDAS achieves the most compelling balance between performance gains and computational efficiency compared to all evaluated baselines.

Programming languages possess rich semantic information - such as data flow - that is represented by graphs and not available from the surface form of source code. Recent code language models have scaled to billions of parameters, but model source code solely as text tokens while ignoring any other structural information. Conversely, models that do encode structural information of code make modifications to the Transformer architecture, limiting their scale and compatibility with pretrained LLMs. In this work, we take the best of both worlds with GALLa - Graph Aligned Large Language Models. GALLa utilizes graph neural networks and cross-modal alignment technologies to inject the structural information of code into LLMs as an auxiliary task during finetuning. This framework is both model-agnostic and task-agnostic, as it can be applied to any code LLM for any code downstream task, and requires the structural graph data only at training time from a corpus unrelated to the finetuning data, while incurring no cost at inference time over the baseline LLM. Experiments on five code tasks with six different baseline LLMs ranging in size from 350M to 14B validate the effectiveness of GALLa, demonstrating consistent improvement over the baseline, even for powerful models such as LLaMA3 and Qwen2.5-Coder.

Recent advancements in Transformer-based large language models (LLMs) have set new standards in natural language processing. However, the classical softmax attention incurs significant computational costs, leading to a $O(T)$ complexity for per-token generation, where $T$ represents the context length. This work explores reducing LLMs' complexity while maintaining performance by introducing Rodimus and its enhanced version, Rodimus$+$. Rodimus employs an innovative data-dependent tempered selection (DDTS) mechanism within a linear attention-based, purely recurrent framework, achieving significant accuracy while drastically reducing the memory usage typically associated with recurrent models. This method exemplifies semantic compression by maintaining essential input information with fixed-size hidden states. Building on this, Rodimus$+$ combines Rodimus with the innovative Sliding Window Shared-Key Attention (SW-SKA) in a hybrid approach, effectively leveraging the complementary semantic, token, and head compression techniques. Our experiments demonstrate that Rodimus$+$-1.6B, trained on 1 trillion tokens, achieves superior downstream performance against models trained on more tokens, including Qwen2-1.5B and RWKV6-1.6B, underscoring its potential to redefine the accuracy-efficiency balance in LLMs. Model code and pre-trained checkpoints are open-sourced at https://github.com/codefuse-ai/rodimus.

Recent advances in Large Language Models (LLMs) have shown promise in function-level code generation, yet repository-level software engineering tasks remain challenging. Current solutions predominantly rely on proprietary LLM agents, which introduce unpredictability and limit accessibility, raising concerns about data privacy and model customization. This paper investigates whether open-source LLMs can effectively address repository-level tasks without requiring agent-based approaches. We demonstrate this is possible by enabling LLMs to comprehend functions and files within codebases through their semantic information and structural dependencies. To this end, we introduce Code Graph Models (CGMs), which integrate repository code graph structures into the LLM's attention mechanism and map node attributes to the LLM's input space using a specialized adapter. When combined with an agentless graph RAG framework, our approach achieves a 43.00% resolution rate on the SWE-bench Lite benchmark using the open-source Qwen2.5-72B model. This performance ranks first among open weight models, second among methods with open-source systems, and eighth overall, surpassing the previous best open-source model-based method by 12.33%.

The key challenge in semantic search is to create models that are both accurate and efficient in pinpointing relevant sentences for queries.While BERT-style bi-encoders excel in efficiency with pre-computed embeddings, they often miss subtle nuances in search tasks.Conversely, GPT-style LLMs with crossencoder designs capture these nuances but are computationally intensive, hindering realtime applications.In this paper, we present D2LLMs-Decomposed and Distilled LLMs for semantic search-that combines the best of both worlds.We decompose a cross-encoder into an efficient bi-encoder integrated with Pooling by Multihead Attention and an Interaction Emulation Module, achieving nuanced understanding and pre-computability.Knowledge from the LLM is distilled into this model using contrastive, rank, and feature imitation techniques.Our experiments show that D2LLM surpasses five leading baselines in terms of all metrics across three tasks, particularly improving NLI task performance by at least 6.45%.

Multi-task learning (MTL) benefits the finetuning of large language models (LLMs) by providing a single model with improved performance and generalization ability across tasks, presenting a resource-efficient alternative to developing separate models for each task.Yet, existing MTL strategies for LLMs often fall short by either being computationally intensive or failing to ensure simultaneous task convergence.This paper presents CoBa, a new MTL approach designed to effectively manage task convergence balance with minimal computational overhead.Utilizing Relative Convergence Scores (RCS), Absolute Convergence Scores (ACS), and a Divergence Factor (DF), CoBa dynamically adjusts task weights during the training process, ensuring that the validation loss of all tasks progress towards convergence at an even pace while mitigating the issue of individual task divergence.The results of our experiments involving four disparate datasets underscore that this approach not only fosters equilibrium in task improvement but enhances the LLMs' performance by up to 13% relative to the secondbest baselines.Code is open-sourced at https: //github.com/codefuse-ai/MFTCoder.

Periodic patterns are a fundamental characteristic of time series in natural world, with significant implications for a range of disciplines, from economics to cloud systems. However, the current literature on periodicity detection faces two key challenges: limited robustness in real-world scenarios and a lack of memory to leverage previously observed time series to accelerate and improve inference on new data. To overcome these obstacles, this paper presents AmortizedPeriod, an innovative approach to periodicity identification based on amortized variational inference that integrates Bayesian statistics and deep learning. Through the Bayesian generative process, our method flexibly captures the dependencies of the periods, trends, noise, and outliers in time series, while also considering missing data and irregular periods in a robust manner. In addition, it utilizes the evidence lower bound of the log-likelihood of the observed time series as the loss function to train a deep attention inference network, facilitating knowledge transfer from the seen time series (and their labels) to unseen ones. Experimental results show that AmortizedPeriod surpasses the state-of-the-art methods by a large margin of 28.5% on average in terms of micro $F_1$-score, with at least 55% less inference time.

Learned lossless data compression has garnered significant attention recently due to its superior compression ratios compared to traditional compressors. However, the computational efficiency of these models jeopardizes their practicality. This paper proposes a novel system for improving the compression ratio while maintaining computational efficiency for learned lossless data compression. Our approach incorporates two essential innovations. First, we propose the Finite-State AutoRegressive (FSAR) entropy coder, an efficient autoregressive Markov model based entropy coder that utilizes a lookup table to expedite autoregressive entropy coding. Next, we present a Straight-Through Hardmax Quantization (STHQ) scheme to enhance the optimization of discrete latent space. Our experiments show that the proposed lossless compression method could improve the compression ratio by up to 6\% compared to the baseline, with negligible extra computational time. Our work provides valuable insights into enhancing the computational efficiency of learned lossless data compression, which can have practical applications in various fields. Code is available at https://github.com/alipay/Finite_State_Autoregressive_Entropy_Coding.

Current directed graph embedding methods build upon undirected techniques but often inadequately capture directed edge information, leading to challenges such as: (1) Suboptimal representations for nodes with low in/out-degrees, due to the insufficient neighbor interactions; (2) Limited inductive ability for representing new nodes post-training; (3) Narrow generalizability, as training is overly coupled with specific tasks. In response, we propose DUPLEX, an inductive framework for complex embeddings of directed graphs. It (1) leverages Hermitian adjacency matrix decomposition for comprehensive neighbor integration, (2) employs a dual GAT encoder for directional neighbor modeling, and (3) features two parameter-free decoders to decouple training from particular tasks. DUPLEX outperforms state-of-the-art models, especially for nodes with sparse connectivity, and demonstrates robust inductive capability and adaptability across various tasks. The code will be available upon publication.

Code LLMs have emerged as a specialized research field, with remarkable studies dedicated to enhancing model's coding capabilities through fine-tuning on pre-trained models. Previous fine-tuning approaches were typically tailored to specific downstream tasks or scenarios, which meant separate fine-tuning for each task, requiring extensive training resources and posing challenges in terms of deployment and maintenance. Furthermore, these approaches failed to leverage the inherent interconnectedness among different code-related tasks. To overcome these limitations, we present a multi-task fine-tuning framework, MFTCoder, that enables simultaneous and parallel fine-tuning on multiple tasks. By incorporating various loss functions, we effectively address common challenges in multi-task learning, such as data imbalance, varying difficulty levels, and inconsistent convergence speeds. Extensive experiments have conclusively demonstrated that our multi-task fine-tuning approach outperforms both individual fine-tuning on single tasks and fine-tuning on a mixed ensemble of tasks. Moreover, MFTCoder offers efficient training capabilities, including efficient data tokenization modes and parameter efficient fine-tuning (PEFT) techniques, resulting in significantly improved speed compared to traditional fine-tuning methods. MFTCoder seamlessly integrates with several mainstream open-source LLMs, such as CodeLLama and Qwen. Our MFTCoder fine-tuned CodeFuse-DeepSeek-33B claimed the top spot on the Big Code Models Leaderboard ranked by WinRate as of January 30, 2024. MFTCoder is open-sourced at https://github.com/codefuse-ai/MFTCOder

Microservice architecture has sprung up over recent years for managing enterprise applications, due to its ability to independently deploy and scale services. Despite its benefits, ensuring the reliability and safety of a microservice system remains highly challenging. Existing anomaly detection algorithms based on a single data modality (i.e., metrics, logs, or traces) fail to fully account for the complex correlations and interactions between different modalities, leading to false negatives and false alarms, whereas incorporating more data modalities can offer opportunities for further performance gain. As a fresh attempt, we propose in this paper a semi-supervised graph-based anomaly detection method, MSTGAD, which seamlessly integrates all available data modalities via attentive multi-modal learning. First, we extract and normalize features from the three modalities, and further integrate them using a graph, namely MST (microservice system twin) graph, where each node represents a service instance and the edge indicates the scheduling relationship between different service instances. The MST graph provides a virtual representation of the status and scheduling relationships among service instances of a real-world microservice system. Second, we construct a transformer-based neural network with both spatial and temporal attention mechanisms to model the inter-correlations between different modalities and temporal dependencies between the data points. This enables us to detect anomalies automatically and accurately in real-time. Extensive experiments on two real-world datasets verify the effectiveness of our proposed MSTGAD method, achieving competitive performance against state-of-the-art approaches, with a 0.961 F1-score and an average increase of 4.85%. The source code of MST-GAD is publicly available at https://github.com/ant-research/microservice_system_twin_graph_based_anomaly_detection.

Bases have become an integral part of modern deep learning-based models for time series forecasting due to their ability to act as feature extractors or future references. To be effective, a basis must be tailored to the specific set of time series data and exhibit distinct correlation with each time series within the set. However, current state-of-the-art methods are limited in their ability to satisfy both of these requirements simultaneously. To address this challenge, we propose BasisFormer, an end-to-end time series forecasting architecture that leverages learnable and interpretable bases. This architecture comprises three components: First, we acquire bases through adaptive self-supervised learning, which treats the historical and future sections of the time series as two distinct views and employs contrastive learning. Next, we design a Coef module that calculates the similarity coefficients between the time series and bases in the historical view via bidirectional cross-attention. Finally, we present a Forecast module that selects and consolidates the bases in the future view based on the similarity coefficients, resulting in accurate future predictions. Through extensive experiments on six datasets, we demonstrate that BasisFormer outperforms previous state-of-the-art methods by 11.04% and 15.78% respectively for univariate and multivariate forecasting tasks. Code is available at: https://github.com/nzl5116190/Basisformer.

Root Cause Analysis (RCA) plays an indispensable role in distributed data system maintenance and operations, as it bridges the gap between fault detection and system recovery. Existing works mainly study multidimensional localization or graph-based root cause localization. This paper opens up the possibilities of exploiting the recently developed framework of explainable AI (XAI) for the purpose of RCA. In particular, we propose BALANCE (BAyesian Linear AttributioN for root CausE localization), which formulates the problem of RCA through the lens of attribution in XAI and seeks to explain the anomalies in the target KPIs by the behavior of the candidate root causes. BALANCE consists of three innovative components. First, we propose a Bayesian multicollinear feature selection (BMFS) model to predict the target KPIs given the candidate root causes in a forward manner while promoting sparsity and concurrently paying attention to the correlation between the candidate root causes. Second, we introduce attribution analysis to compute the attribution score for each candidate in a backward manner. Third, we merge the estimated root causes related to each KPI if there are multiple KPIs. We extensively evaluate the proposed BALANCE method on one synthesis dataset as well as three real-world RCA tasks, that is, bad SQL localization, container fault localization, and fault type diagnosis for Exathlon. Results show that BALANCE outperforms the state-of-the-art (SOTA) methods in terms of accuracy with the least amount of running time, and achieves at least 6% notably higher accuracy than SOTA methods for real tasks. BALANCE has been deployed to production to tackle real-world RCA problems, and the online results further advocate its usage for real-time diagnosis in distributed data systems.

Estimating a sequence of dynamic undirected graphical models, in which adjacent graphs share similar structures, is of paramount importance in various social, financial, biological, and engineering systems, since the evolution of such networks can be utilized for example to spot trends, detect anomalies, predict vulnerability, and evaluate the impact of interventions. Existing methods for learning dynamic graphical models require the tuning parameters that control the graph sparsity and the temporal smoothness to be selected via brute-force grid search. Furthermore, these methods are computationally burdensome with time complexity$\mathcal {O}(NP^3)$for$P$variables and$N$time points. As a remedy, we propose a low-complexity tuning-free Bayesian approach, named BASS. Specifically, we impose temporally dependent spike and slab priors on the graphs such that they are sparse and varying smoothly across time. An efficient variational inference algorithm based on natural gradients is then derived to learn the graph structures from the data in an automatic manner. Owing to the pseudo-likelihood and the mean-field approximation, the time complexity of BASS is only$\mathcal {O}(NP^2)$. To cope with the local maxima problem of variational inference, we resort to simulated annealing and propose a method based on bootstrapping of the observations to generate the annealing noise. We provide numerical evidence that BASS outperforms existing methods on synthetic data in terms of structure estimation, while being more efficient especially when the dimension$P$becomes high. We further apply the approach to the stock return data of 78 banks from 2005 to 2013 and find that the number of edges in the financial network as a function of time contains three peaks, in coincidence with the 2008 global financial crisis and the two subsequent European debt crisis. On the other hand, by identifying the frequency-domain resemblance to the time-varying graphical models, we show that BASS can be extended to learning frequency-varying inverse spectral density matrices, and further yields graphical models for multivariate stationary time series. As an illustration, we analyze scalp EEG signals of patients at the early stages of Alzheimer’s disease (AD) and show that the brain networks extracted by BASS can better distinguish between the patients and the healthy controls.

Smooth handling of pedestrian interactions is a key requirement for Autonomous Vehicles (AV) and Advanced Driver Assistance Systems (ADAS). Such systems call for early and accurate prediction of a pedestrian’s crossing/not-crossing behaviour in front of the vehicle. Existing approaches to pedestrian behaviour prediction make use of pedestrian motion, his/her location in a scene and static context variables such as traffic lights, zebra crossings etc. We stress on the necessity of early prediction for smooth operation of such systems. We introduce the influence of vehicle interactions on pedestrian intention for this purpose. In this paper, we show a discernible advance in prediction time aided by the inclusion of such vehicle interaction context. We apply our methods to two different datasets, one in-house collected - NTU dataset and another public real-life benchmark - JAAD dataset. We also propose a generalization of the Latent-Dynamic Conditional Random Fields (LDCRF), called Factored LDCRF (FLDCRF), for improved sequence prediction performance. FLDCRF outperforms Long Short-Term Memory (LSTM) networks across the datasets over identical time-series features. While the existing best system predicts pedestrian stopping behaviour with 70% accuracy 0.38 seconds before the actual events, our system achieves such accuracy at least 0.9 seconds on an average before the actual events across datasets.

The log-determinant optimization problem with general matrix constraints arises in many applications. The log-determinant term hampers the scalability of existing methods. This paper proposes a highly efficient stochastic method that has time complexity O(N2), whereas existing methods have complexity O(N3). In order to achieve the quadratic complexity, the proposed algorithm leverages an efficient stochastic gradient of the augmented Lagrangian form and relies on subgradient descent method. Convergence of this method is analyzed both theoretically and empirically. The resulting primal-dual stochastic subgradient method yields the same accuracy as existing methods yet only requires O(N2) operations.

Structure learning of Gaussian graphical models typically involves careful tuning of penalty parameters, which balance the tradeoff between data fidelity and graph sparsity. Unfortunately, this tuning is often a “black art” requiring expert experience or brute-force search. It is therefore tempting to develop tuning-free algorithms that can determine the sparsity of the graph adaptively from the observed data in an automatic fashion. In this paper, we propose a novel approach, named BISN (Bayesian inference of Sparse Networks), for automatic Gaussian graphical model selection. Specifically, we regard the off-diagonal entries in the precision matrix as random variables and impose sparse-promoting horseshoe priors on them, resulting in automatic sparsity determination. With the help of stochastic gradients, an efficient variational Bayes algorithm is derived to learn the model. We further propose a decaying recursive stochastic gradient (DRSG) method to reduce the variance of the stochastic gradients and to accelerate the convergence. Our theoretical analysis shows that the time complexity of BISN scales only quadratically with the dimension, whereas the theoretical time complexity of the state-of-the-art methods for automatic graphical model selection is typically a third-order function of the dimension. Furthermore, numerical results show that BISN can achieve comparable or better performance than the state-of-the-art methods in terms of structure recovery, and yet its computational time is several orders of magnitude shorter, especially for large dimensions.

We consider the structure learning problem of Gaussian graphical models when the underlying graph is semi-sparse. More specifically, we assume that the number of edges in the graph grows quadratically with the dimension P. Similar to the case of sparse graphs, the problem is formulated as maximizing the data log-likelihood with an ℓ1norm penalty on the precision matrix (the inverse covariance matrix) that promotes sparsity. We notice that the time complexity of all existing methods is at least O(P3) under the scenario of semi-sparse graphs, thus severely hindering their applications to high-dimensional data. By contrast, the time complexity of the proposed method is only O(P2) with the help of stochastic gradients. We prove the convergence of the proposed algorithm. Numerical results show that the computational time of the proposed method is shorter than that of the state-of-the-art methods when the graph is semi-sparse.

Point set registration presents unique significance in Lidar-based intelligent vehicle localization and mapping. It involves registering point sets of the same scene observed from different positions by determining their relative spatial transformation. However, due to the noise and outliers in the point sets and initial misalignment, existing methods suffer from the issues of low accuracy or large computational cost. In this paper, we propose a novel Bayesian state space model to describe the sequential point registration problem. Specifically, we specify the transformations to be the latent states and further assume that they vary smoothly across time. The point clouds are then represented as Gaussian mixture models that change accordingly with the transformation. We then develop a stochastic variational Bayesian inference algorithm to learning the distributions of the transformation, which automatically strike a balance between mapping every two consecutive point clouds and the temporal smoothness of the transformation. Experimental results based simulated data show that the proposed variational Bayesian point set registration (VB-PSR) algorithm achieves higher accuracy with comparable or less time and resources, in comparison with the state- of-the-art methods.

The simultaneous localization and mapping (SLAM) problem involves using the measurements of sensors to construct an environmental map, while simultaneously recovering the vehicle trajectory within this map. There are broadly two strategies for SLAM: on-line and off-line. In this paper, we focus on the off-line SLAM (a.k.a. full SLAM) problem and propose a variational Bayes inference algorithm to address it. Specifically, the intractable posterior distribution of the vehicle poses given the measurements is approximated by a tractable variational distribution, resulting in estimates of the vehicle poses as well as their uncertainties. In contrast with the existing off- line methods, the inverse variances of the additive noise are updated along with the posterior distribution instead of being fixed, thus, the proposed method is robust to unknown noises. Furthermore, the computational complexity of the proposed method is only linear in the number of frames and the computational bottleneck of the algorithm can be easily parallelized to achieve further acceleration. Numerical results show that the proposed method is insensitive to the selection of the noise parameters. More importantly, it is superior in efficiency to the state-of-the-art method, especially for large- scale SLAM problems.

Dynamic compressed sensing (DCS) has recently gained popularity as a successful approach to recovering dynamic sparse signals. In this paper, we attack the problem from a Bayesian perspective. The proposed model imposes sparse constraints on both the unknown sparse signal and its temporal innovation via t priors. Due to the conjugacy between the priors and likelihoods, we are able to propose a computationally efficient mean-field variational Bayes algorithm to learn the model without parameter tuning. We consider both the online and offline scenarios, and demonstrate via numerical experiments that the proposed methods are superior to alternatives in terms of both reconstruction accuracy and computational time.

We consider the problem of jointly describing extreme events at a multitude of locations, which presents paramount importance in catastrophes forecast and risk management. Specifically, a novel Ensemble-of-Latent-Trees of Pairwise Copula (ELTPC) model is proposed. In this model, the spatial dependence is captured by latent trees expressed by pairwise copulas. To compensate the limited expressiveness of every single latent tree, an mixture of latent trees is employed. By harnessing the variational inference and stochastic gradient techniques, we further develop a triply stochastic variational inference (TSVI) algorithm for learning and inference. The corresponding computational complexity is only linear in the number of variables. Numerical results from both the synthetic and real data show that the ELTPC model provides a reliable description of the spatial extremes in a flexible but parsimonious manner.

A star tracker is a critical sensor for determining and controlling the attitude of a satellite. It utilizes a complementary metal–oxide–semiconductor (CMOS)-active pixel sensor to map the star field onto the focal plane. Starlight is measured and star centroids are calculated to estimate attitude knowledge. In this paper, we present a CMOS image sensor for star centroid measurement in star trackers. To improve sensitivity to low-level starlight, the capacitive transimpedance amplifier pixel is used as the detector. To improve centroiding accuracy, the proposed sensor architecture allows star pixels, pixels that are above a star threshold, to cluster together. The mean value of all the pixels in this cluster is calculated. The star signals are then amplified in relation to this mean value. This increases the signal-to-noise ratio in star regions in line with their starlight intensity. An adaptive region-of-interest readout architecture is also proposed, which reports only star regions instead of the entire frame. The proof-of-concept chip, containing a$128 \times 128$pixel array, was fabricated using AMS 0.35-$\mu \text{m}$CMOS Opto process. Each pixel has a size of$31.2 \times 31.2~\mu \text{m}^{2}$. The measurement results show that centroiding accuracy increases with higher centroiding gain. Within a limited exposure time, the relative centroiding accuracy can surpass that of a commercial image sensor by more than 1%.

This paper presents an antivibration time-delay integration (TDI) CMOS image sensor (CIS) for small remote imaging systems, introducing a hardware-implemented online deblurring (ODB) algorithm to address the image blur problems caused by vibrations. The proposed sensor has eight TDI stages, column-parallel TDI accumulating and ODB circuits. A$256\times 8$-pixel prototype chip was fabricated using a 0.18-$\mu \text{m}$CIS technology with a pixel footprint of$6.5~\mu {\mathrm{ m}}\times 6.5~\mu \text{m}$and a fill factor of 28%. Measurement results show that the sensor can achieve dynamic ranges of 45.1 and 51.8 dB, respectively, with and without enabling the ODB algorithm. Compared with a single-stage line scanner imager, it offers an improvement in signal-to-noise ratios of 1.9 and 8.6 dB, respectively, with and without the ODB algorithm.

Multiscale (multiresolution) graphical models have gained widespread popularity in recent years, since they enjoy rich modeling power as well as efficient inference procedures. Existing approaches to learning multiscale graphical models often leverage the framework of penalized likelihood, and therefore suffer from the issue of regularization selection. In this paper, we propose a novel method to learn multiscale graphical models from the Bayesian perspective. More specifically, the regularization parameters are treated as random variables that follow Gamma distributions. We then derive an efficient variational Bayes algorithm to learn the model, and further demonstrate the advantages of the proposed method through numerical experiments.

Machine vision applications involving the assistance of robots for scene mapping or object classification encounter issues when faced with smooth transparent surfaces, such as glass. Specular reflection from such surfaces saturate the image sensor pixels, restricting their vision of objects beyond the surface. The problem is aggravated by the movement of the robot, making traditional approaches unfeasible. We propose a solution that uses on-chip polarizers to limit the specular reflection and improve scene visibility while overcoming the limited SNR and motion artifacts by using a time delay integration image sensor. We have fabricated a 256×8×5 prototype sensor using a 0.18μm CIS process which achieves a DR of 52.3dB while providing an SNR improvement of 8.8dB over a single-stage linear scanner.

Measurement faults in control systems may result in permanent damages to the system components. Therefore, sensor validation is essential before the measurements are used for any system reconfiguration. In this paper, a statistical approach for sensor fault identification is proposed. Specifically, the potential sensor fault is assumed to be an additive bias term in the measurement model. The problem of fault identification is formulated as a least-squares optimization problem with an ℓ1penalty on the bias term. An algorithm is further introduced to determine the regularization parameter automatically. Experimental results show that the proposed method can accurately detect multiple sensor failures from noisy measurements.

To assess the risk of extreme events such as hurricanes and floods, it is crucial to develop accurate extreme-value statistical models. Extreme events often display heterogeneity, varying continuously with a number of covariates. Previous studies have suggested that models considering covariate effects lead to reliable estimates of extreme value distributions. In this paper, we develop a novel model to incorporate the effects of multiple covariates. Specifically, we analyze as an example the extreme sea states in the Gulf of Mexico, where the distribution of extreme wave heights changes systematically with location and wind direction. The block maxima at each location and sector of wind direction are assumed to follow the Generalized Extreme Value (GEV) distribution. The GEV parameters are coupled across the spatio-directional domain through a graphical model, particularly, a multidimensional thin-membrane model. Efficient learning and inference algorithms are then developed based on the special characteristics of the thin-membrane model. Numerical results for both synthetic and real data indicate that the proposed model can accurately describe marginal behavior of extreme events.

Graphical models are powerful tools to describe complex systems. Especially sparse graphical models are currently en vogue, as they allow us to infer network structure from multiple time series (e.g., functional brain networks from multichannel electroencephalograms). So far, most of the literature deals with stationary time series, whereas real-life time series often exhibit non-stationarity. In this paper, techniques are proposed to infer graphical models from piecewise stationary time series; first change point are detected in the time series, and then graphical models are inferred for each stationary segment. Specifically, a low-complexity algorithm based on Pruned Exact Linear Time method is proposed to identify change points. Copula Gaussian graphical models (with and without hidden variables) are then generated for each stationary segment. The crux of the proposed approach is that it determines the number and location of the change points as well as the graphical models in a fully automated manner. Results for both synthetic data and scalp electroencephalograms of epileptic seizure patients are provided to validate the model.

Assessing the risk of extreme events in a spatial domain, such as hurricanes, floods and droughts, presents unique significance in practice. Unfortunately, the existing extreme-value statistical models are typically not feasible for practical large-scale problems. Graphical models are capable of handling enormous number of variables, yet have not been explored in the realm of extreme-value analysis. To bridge the gap, an extreme-value graphical model is introduced in this paper, i.e., ensemble-of-trees of pairwise copulas (ETPC). In the proposed graphical model, extreme-value marginal distributions are stitched together by means of pairwise copulas, which in turn are the building blocks of the ensemble of trees. By exploiting this particular structure, novel efficient inference algorithms are derived that are applicable to large-scale statistical problems involving extreme values. It is proven that, under mild conditions, the ETPC model exhibits the favorable property of tail-dependence between an arbitrary pair of sites (variables), and therefore is reliable to capture the dependence between extremes at different sites. Real data results further demonstrate the advantages of the ETPC model.

We propose a novel statistical model to describe spatio-temporal extreme events. The model can be used to estimate extreme-value temporal pattern such as seasonality and trend, and further to predict the distribution of extreme events in the future. The basic idea is to explore graphical models to capture the highly structured dependencies among extreme events measured in time and space. More explicitly, we first assume the single observation at each location and time point follows a Generalized Extreme Value (GEV) distribution. The spatio-temporal dependencies are further encoded via graphical models imposed on the GEV parameters. We develop efficient learning and inference algorithms for the resulting non-Gaussian graphical model. Results of both synthetic and real data demonstrate the effectiveness of the proposed approach.

Copula Gaussian graphical models are capable of describing dependencies between a large number of heterogeneous variables. In this paper, low-complexity algorithms are proposed for learning copula Gaussian graphical models from discrete data. The proposed approach is Monte-Carlo expectation maximization: in the E-step, an efficient Gibbs sampler is applied, and in the M-step, the sparse graphical model is inferred by solving a penalized maximize likelihood problem. The regularization parameter is determined through the BINCO method proposed by Li et al. Numerical results for both synthetic and real data demonstrate the effectiveness of the proposed approach.

Gaussian hidden variable graphical models are powerful tools to describe high-dimensional data; they capture dependencies between observed (Gaussian) variables by introducing a suitable number of hidden variables. However, such models are only applicable to Gaussian data. Moreover, they are sensitive to the choice of certain regularization parameters. In this paper, (1) copula Gaussian hidden variable graphical models are introduced, which extend Gaussian hidden variable graphical models to non-Gaussian data; (2) the sparsity pattern of the hidden variable graphical model is learned via stability selection, which leads to more stable results than cross-validation and other methods to select the regularization parameters. The proposed methods are validated on synthetic and real data.

In this paper, we report a novel Time-Delay-Integration (TDI) CMOS image sensor for low-earth orbit (LEO) nano-satellite imaging application, where limited exposure time and unexpected flight fluctuations are major design challenges. The sensor features programmable integration time per stage, dynamic charge transfer path and tunable well capacity. A prototype chip of 1536×8 pixels was implemented using TSMC 0.18µm CMOS image sensor process. Photodiode and other transistors are floor-planned in different arrays, providing small pixel pitch of 3.25µm and high fill factor of 57%.

A novel spatial model for extreme events is proposed. The model may for instance be used to describe the occurrence of catastrophic events such as earthquakes, floods, or hurricanes in certain regions; it may therefore be relevant for, e.g., weather forecasting, urban planning, and environmental assessment. The model is derived from the following ideas: The above-threshold values at each location are assumed to follow a generalized Pareto (GP) distribution. The GP parameters are coupled across space through Markov random fields, in particular, thin-membrane models. The latter are inferred through an empirical Bayes approach. Numerical results are presented for synthetic and real data (related to hurricanes in the Gulf of Mexico).