Large language model (LLM) personalization aims to adapt general-purpose models to individual users.Most existing methods, however, are developed under data-rich and resource-abundant settings, often incurring privacy risks.In contrast, realistic personalization typically occurs after deployment under (i) extremely limited user data, (ii) constrained computational resources, and (iii) strict privacy requirements.We propose PRISP (Privacy-Safe Few-Shot Personalization), a lightweight personalization framework tailored to these constraints.PRISP leverages a Text-to-LoRA hypernetwork to generate task-aware LoRA parameters from task descriptions, and enables efficient user personalization by optimizing a small subset of task-aware LoRA parameters together with minimal additional modules using few-shot user data.Experiments on a few-shot variant of the LaMP benchmark demonstrate that PRISP achieves strong overall performance compared to prior approaches, while reducing computational overhead and eliminating privacy risks.

Domain-Adaptive Pre-training (DAP) has recently gained attention for its effectiveness in fine-tuning pre-trained models. Building on this, continual DAP has been explored to develop pre-trained models capable of incrementally incorporating different domain datasets. However, existing continual DAP methods face several limitations: (1) high computational cost and GPU memory usage during training; (2) sensitivity to incremental data order; and (3) providing a single, generalized model for all end tasks, which contradicts the essence of DAP. In this paper, we propose DoMIX, a novel approach that addresses these challenges by leveraging LoRA modules, a representative parameter-efficient fine-tuning (PEFT) method. Our approach enables efficient and parallel domain-adaptive pre-training that is robust to domain order and effectively utilizes accumulated knowledge to provide tailored pre-trained models for specific tasks.We also demonstrate that our method can be extended beyond the DAP setting to standard LLM fine-tuning scenarios. Code is available at https://github.com/dohoonkim-ai/DoMIX.

Text-to-image (T2I) generative models can create vivid, realistic images from textual descriptions. As these models proliferate, they expose new concerns about their ability to represent diverse demographic groups, propagate stereo-types, and efface minority populations. Despite growing attention to the "safe" and "responsible" design of artificial intelligence (AI), there is no established methodology to systematically measure and control representational harms in image generation. This paper introduces a novel frame-work to measure the representation of intersectional groups in images generated by T2I models by applying the Multi-Group Proportional Representation (MPR) metric. MPR evaluates the worst-case deviation of representation statistics across given population groups in images produced by a generative model, allowing for flexible and context-specific measurements based on user requirements. We also develop an algorithm to optimize T2I models for this metric. Through experiments, we demonstrate that MPR can effectively mea-sure representation statistics across multiple intersectional groups and, when used as a training objective, can guide models toward a more balanced generation across demo-graphic groups while maintaining generation quality.1

We argue that the negative transfer problem occurring when the new task to learn arrives is an important problem that needs not be overlooked when developing effective Continual Reinforcement Learning (CRL) algorithms. Through comprehensive experimental validation, we demonstrate that such issue frequently exists in CRL and cannot be effectively addressed by several recent work on either mitigating plasticity loss of RL agents or enhancing the positive transfer in CRL scenario. To that end, we develop Reset & Distill (R&D), a simple yet highly effective baseline method, to overcome the negative transfer problem in CRL. R&D combines a strategy of resetting the agent's online actor and critic networks to learn a new task and an offline learning step for distilling the knowledge from the online actor and previous expert's action probabilities. We carried out extensive experiments on long sequence of Meta World tasks and show that our simple baseline method consistently outperforms recent approaches, achieving significantly higher success rates across a range of tasks. Our findings highlight the importance of considering negative transfer in CRL and emphasize the need for robust strategies like R&D to mitigate its detrimental effects.

Offline goal-conditioned reinforcement learning (GCRL) offers a practical learning paradigm in which goal-reaching policies are trained from abundant state–action trajectory datasets without additional environment interaction. However, offline GCRL still struggles with long-horizon tasks, even with recent advances that employ hierarchical policy structures, such as HIQL. Identifying the root cause of this challenge, we observe the following insight. Firstly, performance bottlenecks mainly stem from the high-level policy’s inability to generate appropriate subgoals. Secondly, when learning the high-level policy in the long-horizon regime, the sign of the advantage estimate frequently becomes incorrect. Thus, we argue that improving the value function to produce a clear advantage estimate for learning the high-level policy is essential. In this paper, we propose a simple yet effective solution: _**Option-aware Temporally Abstracted**_ value learning, dubbed **OTA**, which incorporates temporal abstraction into the temporal-difference learning process. By modifying the value update to be _option-aware_, our approach contracts the effective horizon length, enabling better advantage estimates even in long-horizon regimes. We experimentally show that the high-level policy learned using the OTA value function achieves strong performance on complex tasks from OGBench, a recently proposed offline GCRL benchmark, including maze navigation and visual robotic manipulation environments. Our code is available at https://github.com/ota-v/ota-v

An image classifier may depend on incidental features stemming from a strong correlation between the feature and the classification target in the training dataset. Recently, Last Layer Retraining (LLR) with group-balanced datasets is shown to be efficient in mitigating the spurious correlation of classifiers. However, the acquisition of image-based group-balanced datasets is costly, which hinders the general applicability of the LLR method. In this work, we propose to perform LLR based on text datasets built with large language models to debias a general image classifier. To that end, we demonstrate that text can generally be a proxy for its corresponding image beyond the image-text joint embedding space, which is achieved with a linear projector that ensures orthogonality between its weight and the modality gap of the joint embedding space. In addition, we propose a systematic validation procedure that checks whether the generated words are compatible with the embedding space of CLIP and the image classifier, which is shown to be effective for improving debiasing performance. We dub these procedures as TLDR (Text-based Last layer retraining for Debiasing image classifieRs) and show our method achieves the performance that is competitive with the LLR methods that require group-balanced image dataset for retraining. Furthermore, TLDR outperforms other baselines that involve training the last layer without any group annotated dataset. Codes: https://github.com/beotborry/TLDR

Although significant advances have been made in both numerical methods and machine learning approaches for solving differential equations across various scientific sectors, these methods often rely on complete information about the differential equations, including precise parameter values, which are not easily obtainable in real-world scenarios. To address this challenge, data-driven methods for discovering differential equations have gained growing popularity in recent years. However, many existing approaches demand unrealistic prerequisites, such as extensive high-fidelity data or carefully designed low-fidelity data and functions. In this paper, we propose a novel method: P hysics- I nformed F ine- T uning ( PIFT ) to discover the unknown parameters in differential equations when only randomly distributed sparse data points are available. PIFT consists of three stages; (i) generating low fidelity data from prior knowledge under realistic settings, (ii) pre-training a single neural network with the generated low fidelity data, and (iii) fine-tuning the pre-trained model using physics-informed loss function. PIFT is evaluated on seven scientific problems including five ordinary differential equations and two partial differential equations. We also demonstrate the robustness and generalizability of PIFT to out-of-distribution tasks. PIFT exhibits high accuracy and robustness in discovering unknown parameters of differential equations from randomly distributed sparse data points.

Since the recent advent of regulations for data protection (e.g., the General Data Protection Regulation), there has been increasing demand in deleting information learned from sensitive data in pre-trained models without retraining from scratch. The inherent vulnerability of neural networks towards adversarial attacks and unfairness also calls for a robust method to remove or correct information in an instance-wise fashion, while retaining the predictive performance across remaining data. To this end, we consider instance-wise unlearning, of which the goal is to delete information on a set of instances from a pre-trained model, by either misclassifying each instance away from its original prediction or relabeling the instance to a different label. We also propose two methods that reduce forgetting on the remaining data: 1) utilizing adversarial examples to overcome forgetting at the representation-level and 2) leveraging weight importance metrics to pinpoint network parameters guilty of propagating unwanted information. Both methods only require the pre-trained model and data instances to forget, allowing painless application to real-life settings where the entire training set is unavailable. Through extensive experimentation on various image classification benchmarks, we show that our approach effectively preserves knowledge of remaining data while unlearning given instances in both single-task and continual unlearning scenarios.

We consider a critical issue of false negatives in Vision-Language Pre-training (VLP), a challenge that arises from the inherent many-to-many correspondence of image-text pairs in large-scale web-crawled datasets. The presence of false negatives can impede achieving optimal performance and even lead to a significant performance drop. To address this challenge, we propose MAFA (MAnaging FAlse negatives), which consists of two pivotal components building upon the recently developed GRouped mIni-bcTch sampling (GRIT) strategy: 1) an efficient connection mining process that identifies and converts false negatives into positives, and 2) label smoothing for the image-text contrastive (ITC) loss. Our comprehensive experiments verify the effectiveness of MAFA across multiple downstream tasks, emphasizing the crucial role of addressing false negatives in VLP, potentially even surpassing the importance of addressing false positives. In addition, the compatibility of MAFA with the recent BLIP-family model is also demonstrated. Code is available at https://github.com/jaeseokbyun/MAFA.

Most continual learning (CL) algorithms have focused on tackling the stability-plasticity dilemma, that is, the challenge of preventing the forgetting of past tasks while learning new ones. However, we argue that they have overlooked the impact of knowledge transfer when the training dataset of a certain task is biased — namely, when the dataset contains some spurious correlations that can overly influence the prediction rule of a model. In that case, how would the dataset bias of a certain task affect the prediction rules of a CL model for future or past tasks? In this work, we carefully design systematic experiments using three benchmark datasets to answer the question from our empirical findings. Specifically, we first show through two-task CL experiments that standard CL methods, which are oblivious of the dataset bias, can transfer bias from one task to another, both forward and backward. Moreover, we find out this transfer is exacerbated depending on whether the CL methods focus on stability or plasticity. We then present that the bias is also transferred and even accumulates in longer task sequences. Finally, we offer a standardized experimental setup and a simple, yet strong plug-in baseline method, dubbed as group-class Balanced Greedy Sampling (BGS), which are utilized for the development of more advanced bias-aware CL methods.

We introduce a novel Pseudo-Negative Regularization (PNR) framework for effective continual self-supervised learning (CSSL). Our PNR leverages pseudo-negatives obtained through model-based augmentation in a way that newly learned representations may not contradict what has been learned in the past. Specifically, for the InfoNCE-based contrastive learning methods, we define symmetric pseudo-negatives obtained from current and previous models and use them in both main and regularization loss terms. Furthermore, we extend this idea to non-contrastive learning methods which do not inherently rely on negatives. For these methods, a pseudo-negative is defined as the output from the previous model for a differently augmented version of the anchor sample and is asymmetrically applied to the regularization term. Extensive experimental results demonstrate that our PNR framework achieves state-of-the-art performance in representation learning during CSSL by effectively balancing the trade-off between plasticity and stability.

In Reinforcement Learning (RL), designing precise reward functions remains to be a challenge, particularly when aligning with human intent. Preference-based RL (PbRL) was introduced to address this problem by learning reward models from human feedback. However, existing PbRL methods have limitations as they often overlook the second-order preference that indicates the relative strength of preference. In this paper, we propose Listwise Reward Estimation (LiRE), a novel approach for offline PbRL that leverages second-order preference information by constructing a Ranked List of Trajectories (RLT), which can be efficiently built by using the same ternary feedback type as traditional methods. To validate the effectiveness of LiRE, we propose a new offline PbRL dataset that objectively reflects the effect of the estimated rewards. Our extensive experiments on the dataset demonstrate the superiority of LiRE, i.e., outperforming state-of-the-art baselines even with modest feedback budgets and enjoying robustness with respect to the number of feedbacks and feedback noise. Our code is available at https://github.com/chwoong/LiRE

The notion of algorithmic fairness has been actively explored from various aspects of fairness, such as counterfactual fairness (CF) and group fairness (GF). However, the exact relationship between CF and GF remains to be unclear, especially in image classification tasks; the reason is because we often cannot collect counterfactual samples regarding a sensitive attribute, essential for evaluating CF, from the existing images (e.g., a photo of the same person but with different secondary sex characteristics). In this paper, we construct new image datasets for evaluating CF by using a high-quality image editing method and carefully labeling with human annotators. Our datasets, CelebA-CF and LFW-CF, build upon the popular image GF benchmarks; hence, we can evaluate CF and GF simultaneously. We empirically observe that CF does not imply GF in image classification, whereas previous studies on tabular datasets observed the opposite. We theoretically show that it could be due to the existence of a latent attribute $G$ that is correlated with, but not caused by, the sensitive attribute (e.g., secondary sex characteristics are highly correlated with hair length). From this observation, we propose a simple baseline, Counterfactual Knowledge Distillation (CKD), to mitigate such correlation with the sensitive attributes. Extensive experimental results on CelebA-CF and LFW-CF demonstrate that CF-achieving models satisfy GF if we successfully reduce the reliance on $G$ (e.g., using CKD).

We propose a novel and effective purification-based adversarial defense method against pre-processor blind white-and black-box attacks, without requiring any adversarial training or retraining of the classification model. Based on the observation of the adversarial noise, we propose a simple iterative Gaussian Smoothing (GS) that smoothes out adversarial noise and achieves substantially high robust accuracy. To further improve the method, we propose Neural Contextual Iterative Smoothing (NCIS), which trains a blind-spot network (BSN) in a self-supervised manner to reconstruct the discriminative features of the smoothed original image. From the extensive experiments on the large-scale ImageNet, we show that our method achieves both competitive standard accuracy and state-of-the-art robust accuracy against most strong purifier-blind white- and black-box attacks. Also, we propose a new evaluation benchmark based on commercial image classification APIs, including AWS, Azure, Clarifai, and Google, and demonstrate that users can use our method to increase the adversarial robustness of APIs.

Neural network interpretation methods, particularly feature attribution methods, are known to be fragile with respect to adversarial input perturbations. To address this, several methods for enhancing the local smoothness of the gradient while training have been proposed for attaining robust feature attributions. However, the lack of considering the normalization of the attributions, which is essential in their visualizations, has been an obstacle to understanding and improving the robustness of feature attribution methods. In this paper, we provide new insights by taking such normalization into account. First, we show that for every non-negative homogeneous neural network, a naive l2-robust criterion for gradients is not normalization invariant, which means that two functions with the same normalized gradient can have different values. Second, we formulate a normalization invariant cosine distance-based criterion and derive its upper bound, which gives insight for why simply minimizing the Hessian norm at the input, as has been done in previous work, is not sufficient for attaining robust feature attribution. Finally, we propose to combine both l2 and cosine distance-based criteria as regularization terms to leverage the advantages of both in aligning the local gradient. As a result, we experimentally show that models trained with our method produce much more robust interpretations on CIFAR-10 and ImageNet-100 without significantly hurting the accuracy, compared to the recent baselines. To the best of our knowledge, this is the first work to verify the robustness of interpretation on a larger-scale dataset beyond CIFAR-10, thanks to the computational efficiency of our method.

Batch Normalization (BN) and its variants has been extensively studied for neural nets in various computer vision tasks, but relatively little work has been dedicated to studying the effect of BN in continual learning. To that end, we develop a new update patch for BN, particularly tailored for the exemplar-based class-incremental learning (CIL). The main issue of BN in CIL is the imbalance of training data between current and past tasks in a mini-batch, which makes the empirical mean and variance as well as the learnable affine transformation parameters of BN heavily biased toward the current task - contributing to the forgetting of past tasks. While one of the recent BN variants has been developed for” online” CIL; in which the training is done with a single epoch, we show that their method does not necessarily bring gains for “offline” CIL, in which a model is trained with multiple epochs on the imbalanced training data. The main reason for the ineffectiveness of their method lies in not fully addressing the data imbalance issue, especially in computing the gradients for learning the affine transformation parameters of BN. Accordingly, our new hyperparameter-free variant, dubbed as Task-Balanced BN (TBBN), is proposed to more correctly resolve the imbalance issue by making a horizontally-concatenated task-balanced batch using both reshape and repeat operations during training. Based on our experiments on class incremental learning of CIFAR-100, ImageNet-100, and five dissimilar task datasets, we demonstrate that our TBBN, which works exactly the same as the vanilla BN in the inference time, is easily applicable to most existing exemplar-based offline CIL algorithms and consistently outperforms other BN variants.

Modeling spatiotemporal brain dynamics from high-dimensional data, such as functional Magnetic Resonance Imaging (fMRI), is a formidable task in neuroscience. Existing approaches for fMRI analysis utilize hand-crafted features, but the process of feature extraction risks losing essential information in fMRI scans. To address this challenge, we present SwiFT (Swin 4D fMRI Transformer), a Swin Transformer architecture that can learn brain dynamics directly from fMRI volumes in a memory and computation-efficient manner. SwiFT achieves this by implementing a 4D window multi-head self-attention mechanism and absolute positional embeddings. We evaluate SwiFT using multiple large-scale resting-state fMRI datasets, including the Human Connectome Project (HCP), Adolescent Brain Cognitive Development (ABCD), and UK Biobank (UKB) datasets, to predict sex, age, and cognitive intelligence. Our experimental outcomes reveal that SwiFT consistently outperforms recent state-of-the-art models. Furthermore, by leveraging its end-to-end learning capability, we show that contrastive loss-based self-supervised pre-training of SwiFT can enhance performance on downstream tasks. Additionally, we employ an explainable AI method to identify the brain regions associated with sex classification. To our knowledge, SwiFT is the first Swin Transformer architecture to process dimensional spatiotemporal brain functional data in an end-to-end fashion. Our work holds substantial potential in facilitating scalable learning of functional brain imaging in neuroscience research by reducing the hurdles associated with applying Transformer models to high-dimensional fMRI.

Recently, fairness-aware learning have become increasingly crucial, but most of those methods operate by assuming the availability of fully annotated demographic group labels. We emphasize that such assumption is unrealistic for real-world applications since group label annotations are expensive and can conflict with privacy issues. In this paper, we consider a more practical scenario, dubbed as Algorithmic Group Fairness with the Partially annotated Group labels (Fair-PG). We observe that the existing methods to achieve group fairness perform even worse than the vanilla training, which simply uses full data only with target labels, under Fair-PG. To address this problem, we propose a simple Confidence-based Group Label assignment (CGL) strategy that is readily applicable to any fairness-aware learning method. CGL utilizes an auxiliary group classifier to assign pseudo group labels, where random labels are assigned to low confident samples. We first theoretically show that our method design is better than the vanilla pseudo-labeling strategy in terms of fairness criteria. Then, we empirically show on several benchmark datasets that by combining CGL and the state-of-the-art fairness-aware in-processing methods, the target accuracies and the fairness metrics can be jointly improved compared to the baselines. Furthermore, we convincingly show that CGL enables to naturally augment the given group-labeled dataset with external target label-only datasets so that both accuracy and fairness can be improved. Code is available at https://github.com/naver-ai/cgl_fairness.

Heterogeneous graph neural networks (GNNs) achieve strong performance on node classification tasks in a semi-supervised learning setting. However, as in the simpler homogeneous GNN case, message-passing-based heterogeneous GNNs may struggle to balance between resisting the oversmoothing that may occur in deep models, and capturing long-range dependencies of graph structured data. Moreover, the complexity of this trade-off is compounded in the heterogeneous graph case due to the disparate heterophily relationships between nodes of different types. To address these issues, we propose a novel heterogeneous GNN architecture in which layers are derived from optimization steps that descend a novel relation-aware energy function. The corresponding minimizer is fully differentiable with respect to the energy function parameters, such that bilevel optimization can be applied to effectively learn a functional form whose minimum provides optimal node representations for subsequent classification tasks. In particular, this methodology allows us to model diverse heterophily relationships between different node types while avoiding oversmoothing effects. Experimental results on 8 heterogeneous graph benchmarks demonstrates that our proposed method can achieve competitive node classification accuracy.

Human activity classification based on micro-Doppler signatures measured by radar recently has been successfully applied in diverse applications due to the improvement of machine learning methods, e.g., deep convolutional neural networks (DCNNs). Despite the success, those methods encounter a common practical problem when all the radar data are not readily available before training a model but sequentially arrive as the learning continues, e.g., during surveillance or search-and-rescue operations. That is, when DCNN is naively utilized in such settings, it is well-known that catastrophic forgetting of past learned tasks occurs; hence, more robust continual learning methods should be developed. To that end, we apply several state-of-the-art methods for two practical continual learning scenarios in activity classification, i.e., when the data for a subject and an activity class arrive incrementally, respectively, and compare the competitiveness of those methods. To the best of our knowledge, this is the first comparative study of continual learning methods for the classification based on micro-Doppler signatures—we find that exemplar memory-based methods particularly become very effective for both scenarios, not only for the performance but also for the memory usage.

We consider the challenging blind denoising problem for Poisson-Gaussian noise, in which no additional information about clean images or noise level parameters is available. Particularly, when only "single" noisy images are available for training a denoiser, the denoising performance of existing methods was not satisfactory. Recently, the blind pixelwise affine image denoiser (BP-AIDE) was proposed and significantly improved the performance in the above setting, to the extent that it is competitive with denoisers which utilized additional information. However, BP-AIDE seriously suffered from slow inference time due to the inefficiency of noise level estimation procedure and that of the blind-spot network (BSN) architecture it used. To that end, we propose Fast Blind Image Denoiser (FBI-Denoiser) for Poisson-Gaussian noise, which consists of two neural network models; 1) PGE-Net that estimates Poisson-Gaussian noise parameters 2000 times faster than the conventional methods and 2) FBI-Net that realizes a much more efficient BSN for pixelwise affine denoiser in terms of the number of parameters and inference speed. Consequently, we show that our FBI-Denoiser blindly trained solely based on single noisy images can achieve the state-of-the-art performance on several real-world noisy image benchmark datasets with much faster inference time (× 10), compared to BP-AIDE.

We consider class incremental learning (CIL) problem, in which a learning agent continuously learns new classes from incrementally arriving training data batches and aims to predict well on all the classes learned so far. The main challenge of the problem is the catastrophic forgetting, and for the exemplar-memory based CIL methods, it is generally known that the forgetting is commonly caused by the classification score bias that is injected due to the data imbalance between the new classes and the old classes (in the exemplar-memory). While several methods have been proposed to correct such score bias by some additional post-processing, e.g., score re-scaling or balanced fine-tuning, no systematic analysis on the root cause of such bias has been done. To that end, we analyze that computing the softmax probabilities by combining the output scores for all old and new classes could be the main cause of the bias. Then, we propose a new method, dubbed as Separated Softmax for Incremental Learning (SS-IL), that consists of separated softmax (SS) output layer combined with task-wise knowledge distillation (TKD) to resolve such bias. Throughout our extensive experimental results on several large-scale CIL benchmark datasets, we show our SS-IL achieves strong state-of-the-art accuracy through attaining much more balanced prediction scores across old and new classes, without any additional post-processing.

We consider a class-incremental semantic segmentation (CISS) problem. While some recently proposed algorithms utilized variants of knowledge distillation (KD) technique to tackle the problem, they only partially addressed the key additional challenges in CISS that causes the catastrophic forgetting; \textit{i.e.}, the semantic drift of the background class and multi-label prediction issue. To better address these challenges, we propose a new method, dubbed as SSUL-M (Semantic Segmentation with Unknown Label with Memory), by carefully combining several techniques tailored for semantic segmentation. More specifically, we make three main contributions; (1) modeling \textit{unknown} class within the background class to help learning future classes (help plasticity), (2) \textit{freezing} backbone network and past classifiers with binary cross-entropy loss and pseudo-labeling to overcome catastrophic forgetting (help stability), and (3) utilizing \textit{tiny exemplar memory} for the first time in CISS to improve \textit{both} plasticity and stability. As a result, we show our method achieves significantly better performance than the recent state-of-the-art baselines on the standard benchmark datasets. Furthermore, we justify our contributions with thorough and extensive ablation analyses and discuss different natures of the CISS problem compared to the standard class-incremental learning for classification. The official code is available at https://github.com/clovaai/SSUL.

We devise a novel neural network-based universal denoiser for the finite-input, general-output (FIGO) channel. Based on the assumption of known noisy channel densities, which is realistic in many practical scenarios, we train the network such that it can denoise as well as the best sliding window denoiser for any given underlying clean source data. Our algorithm, dubbed as Generalized CUDE (Gen-CUDE), enjoys several desirable properties; it can be trained in an unsupervised manner (solely based on the noisy observation data), has much smaller computational complexity compared to the previously developed universal denoiser for the same setting, and has much tighter upper bound on the denoising performance, which is obtained by a theoretical analysis. In our experiments, we show such tighter upper bound is also realized in practice by showing that Gen-CUDE achieves much better denoising results compared to other strong baselines for both synthetic and real underlying clean sequences.

We propose a novel regularization-based continual learning method, dubbed as Adaptive Group Sparsity based Continual Learning (AGS-CL), using two group sparsity-based penalties. Our method selectively employs the two penalties when learning each neural network node based on its the importance, which is adaptively updated after learning each task. By utilizing the proximal gradient descent method, the exact sparsity and freezing of the model is guaranteed during the learning process, and thus, the learner explicitly controls the model capacity. Furthermore, as a critical detail, we re-initialize the weights associated with unimportant nodes after learning each task in order to facilitate efficient learning and prevent the negative transfer. Throughout the extensive experimental results, we show that our AGS-CL uses orders of magnitude less memory space for storing the regularization parameters, and it significantly outperforms several state-of-the-art baselines on representative benchmarks for both supervised and reinforcement learning.

We propose a novel iterative channel estimation (ICE) algorithm that essentially removes the critical known noisy channel assumption for universal discrete denoising problem. Our algorithm is based on Neural DUDE (N-DUDE), a recently proposed neural network-based discrete denoiser, and it estimates the channel transition matrix as well as the neural network parameters in an alternating manner until convergence. While we do not make any probabilistic assumption on the underlying clean data, our ICE resembles Expectation-Maximization (EM) with variational approximation, and it takes advantage of the property that N-DUDE can always induce a marginal posterior distribution of the clean data. We carefully validate the channel estimation quality of ICE, and with extensive experiments on several radically different types of data, we show the ICE equipped neural network-based denoisers can perform \emph{universally} well regardless of the uncertainties in both the channel and the clean source. Moreover, we show ICE becomes extremely robust to its hyperparameters, and show the denoisers with ICE significantly outperform the strong baseline that can handle the channel uncertainties for denoising, the widely used Baum-Welch (BW) algorithm for hidden Markov models (HMM).

Convolutional neural network (CNN)-based denoisers are recently shown to overwhelm the denoising performances of the conventional prior- or optimization-based methods, particularly for the additive white Gaussian noise (AWGN) case. However, those typical CNN-based denoisers had two main limitations: (1) the superb denoising performances for AWGN do not necessarily carry over to the signal-dependent real-noise cases and (2) the clean source images are required as targets for supervised training. Both limitations are critical in practical real-noise denoising setting, since neither the noise characteristics nor the clean source images are readily available for supervised training. While some recent work proposed CNN denoisers that are trained solely with noisy images, they still fall short of the classical baselines, e.g., BM3D, for blind real-noise denoising settings, in which nothing is known about the noise and only noisy images are available for training. In this paper, we propose Blind Pixelwise Affine Image DEnoiser (BP-AIDE), which tackles above two limitations by combining variance stabilizing transformation (VST) technique with a recently developed unbiased estimate of the mean-squared error (MSE) for a pixelwise affine denoiser. As a result, we show BP-AIDE can be blindly trained solely with single real-noise corrupted images and, to the best of our knowledge, first outperforms both CNN-based and conventional denoisers for the same setting, on several real-world noisy image benchmark datasets.

We propose DoPAMINE, a new neural network based multiplicative noise despeckling algorithm. Our algorithm is inspired by Neural AIDE (N-AIDE), which is a recently proposed neural adaptive image denoiser. While the original NAIDE was designed for the additive noise case, we show that the same framework, i.e., adaptively learning a network for pixel-wise affine denoisers by minimizing an unbiased estimate of MSE, can be applied to the multiplicative noise case as well. Moreover, we derive a double-sided masked CNN architecture which can control the variance of the activation values in each layer and converge fast to high denoising performance during supervised training. In the experimental results, we show our DoPAMINE possesses high adaptivity via fine-tuning the network parameters based on the given noisy image and achieves significantly better despeckling results compared to SAR-DRN, a state-of-the-art CNN-based algorithm.

Non-intrusive load monitoring (NILM), also known as energy disaggregation, is a blind source separation problem where a household’s aggregate electricity consumption is broken down into electricity usages of individual appliances. In this way, the cost and trouble of installing many measurement devices over numerous household appliances can be avoided, and only one device needs to be installed. The problem has been well-known since Hart’s seminal paper in 1992, and recently significant performance improvements have been achieved by adopting deep networks. In this work, we focus on the idea that appliances have on/off states, and develop a deep network for further performance improvements. Specifically, we propose a subtask gated network that combines the main regression network with an on/off classification subtask network. Unlike typical multitask learning algorithms where multiple tasks simply share the network parameters to take advantage of the relevance among tasks, the subtask gated network multiply the main network’s regression output with the subtask’s classification probability. When standby-power is additionally learned, the proposed solution surpasses the state-of-the-art performance for most of the benchmark cases. The subtask gated network can be very effective for any problem that inherently has on/off states.

We propose a new image denoising algorithm, dubbed as Fully Convolutional Adaptive Image DEnoiser (FC-AIDE), that can learn from an offline supervised training set with a fully convolutional neural network as well as adaptively fine-tune the supervised model for each given noisy image. We significantly extend the framework of the recently proposed Neural AIDE, which formulates the denoiser to be context-based pixelwise mappings and utilizes the unbiased estimator of MSE for such denoisers. The two main contributions we make are; 1) implementing a novel fully convolutional architecture that boosts the base supervised model, and 2) introducing regularization methods for the adaptive fine-tuning such that a stronger and more robust adaptivity can be attained. As a result, FC-AIDE is shown to possess many desirable features; it outperforms the recent CNN-based state-of-the-art denoisers on all of the benchmark datasets we tested, and gets particularly strong for various challenging scenarios, e.g., with mismatched image/noise characteristics or with scarce supervised training data. The source code our algorithm is available at https://github.com/csm9493/FC-AIDE-Keras.

We introduce a new neural network-based continual learning algorithm, dubbed as Uncertainty-regularized Continual Learning (UCL), which builds on traditional Bayesian online learning framework with variational inference. We focus on two significant drawbacks of the recently proposed regularization-based methods: a) considerable additional memory cost for determining the per-weight regularization strengths and b) the absence of gracefully forgetting scheme, which can prevent performance degradation in learning new tasks. In this paper, we show UCL can solve these two problems by introducing a fresh interpretation on the Kullback-Leibler (KL) divergence term of the variational lower bound for Gaussian mean-field approximation. Based on the interpretation, we propose the notion of node-wise uncertainty, which drastically reduces the number of additional parameters for implementing per-weight regularization. Moreover, we devise two additional regularization terms that enforce \emph{stability} by freezing important parameters for past tasks and allow \emph{plasticity} by controlling the actively learning parameters for a new task. Through extensive experiments, we show UCL convincingly outperforms most of recent state-of-the-art baselines not only on popular supervised learning benchmarks, but also on challenging lifelong reinforcement learning tasks. The source code of our algorithm is available at https://github.com/csm9493/UCL.

We ask whether the neural network interpretation methods can be fooled via adversarial model manipulation, which is defined as a model fine-tuning step that aims to radically alter the explanations without hurting the accuracy of the original models, e.g., VGG19, ResNet50, and DenseNet121. By incorporating the interpretation results directly in the penalty term of the objective function for fine-tuning, we show that the state-of-the-art saliency map based interpreters, e.g., LRP, Grad-CAM, and SimpleGrad, can be easily fooled with our model manipulation. We propose two types of fooling, Passive and Active, and demonstrate such foolings generalize well to the entire validation set as well as transfer to other interpretation methods. Our results are validated by both visually showing the fooled explanations and reporting quantitative metrics that measure the deviations from the original explanations. We claim that the stability of neural network interpretation method with respect to our adversarial model manipulation is an important criterion to check for developing robust and reliable neural network interpretation method.

We propose a novel neural network-based adaptive image denoiser, dubbased as Neural AIDE. Unlike other neural network-based denoisers, which typically apply supervised training to learn a mapping from a noisy patch to a clean patch, we formulate to train a neural network to learn context-based affine mappings that get applied to each noisy pixel. Our formulation enables using SURE (Stein's Unbiased Risk Estimator)-like estimated losses of those mappings as empirical risks to minimize. In results, we can combine both supervised training of the network parameters from a separate dataset and adaptive fine-tuning of them using the given noisy image subject to denoising. Our algorithm with a plain fully connected architecture is shown to attain a competitive denoising performance on benchmark datasets compared to the strong baselines. Furthermore, Neural AIDE can robustly correct the mismatched noise level in the supervised learning via fine-tuning, of which adaptivity is absent in other neural network-based denoisers.

We present a new framework of applying deep neural networks (DNN) to devise a universal discrete denoiser. Unlike other approaches that utilize supervised learning for denoising, we do not require any additional training data. In such setting, while the ground-truth label, i.e., the clean data, is not available, we devise ``pseudo-labels'' and a novel objective function such that DNN can be trained in a same way as supervised learning to become a discrete denoiser. We experimentally show that our resulting algorithm, dubbed as Neural DUDE, significantly outperforms the previous state-of-the-art in several applications with a systematic rule of choosing the hyperparameter, which is an attractive feature in practice.

We propose the use of deep convolutional neural networks (DCNNs) for human detection and activity classification based on Doppler radar. Previously, proposed schemes for these problems remained in the conventional supervised learning paradigm that relies on the design of handcrafted features. Whereas these schemes attained high accuracy, the requirement for domain knowledge of each problem limits the scalability of the proposed schemes. In this letter, we present an alternative deep learning approach. We apply the DCNN, one of the most successful deep learning algorithms, directly to a raw micro-Doppler spectrogram for both human detection and activity classification problem. The DCNN can jointly learn the necessary features and classification boundaries using the measured data without employing any explicit features on the micro-Doppler signals. We show that the DCNN can achieve accuracy results of 97.6% for human detection and 90.9% for human activity classification.

Recently, recurrent neural networks (RNN) have achieved the state-of-the-art performance in several applications that deal with temporal data, e.g., speech recognition, handwriting recognition and machine translation. While the ability of handling long-term dependency in data is the key for the success of RNN, combating over-fitting in training the models is a critical issue for achieving the cutting-edge performance particularly when the depth and size of the network increase. To that end, there have been some attempts to apply the dropout, a popular regularization scheme for the feed-forward neural networks, to RNNs, but they do not perform as well as other regularization scheme such as weight noise injection. In this paper, we propose rnnDrop, a novel variant of the dropout tailored for RNNs. Unlike the existing methods where dropout is applied only to the non-recurrent connections, the proposed method applies dropout to the recurrent connections as well in such a way that RNNs generalize well. Our experiments show that rnnDrop is a better regularization method than others including weight noise injection. Namely, when deep bidirectional long short-term memory (LSTM) RNNs were trained with rnnDrop as acoustic models for phoneme and speech recognition, they significantly outperformed the current state-of-the-arts; we achieved the phoneme error rate of 16.29% on the TIMIT core test set for phoneme recognition and the word error rate of 5.53% on the Wall Street Journal (WSJ) dataset, dev93, for speech recognition, which are the best reported results on both of the datasets.

Satellite-retrieved aerosol optical depth (AOD) can potentially provide an effective way to complement the spatial coverage limitation of a ground particulate air-pollution monitoring network such as the U.S. Environment Protection Agency's regulatory monitoring network. One of the current state-of-the-art AOD retrieval methods is the National Aeronautics and Space Administration's Multiangle Imaging SpectroRadiometer (MISR) operational algorithm, which has a spatial resolution of 17.6 km × 17.6 km. Although the MISR's aerosol products lead to exciting research opportunities to study particle composition at a regional scale, its spatial resolution is too coarse for analyzing urban areas, where the air pollution has stronger spatial variations and can severely impact public health and the environment. Accordingly, a novel AOD retrieval algorithm with a resolution of 4.4 km × 4.4 km has been recently developed, which is based on hierarchical Bayesian modeling and the Monte Carlo Markov chain (MCMC) inference method. In this paper, we carry out detailed quantitative and qualitative evaluations of the new algorithm, which is called the HB-MCMC algorithm, using recent AErosol RObotic NETwork (AERONET) Distributed Regional Aerosol Gridded Observation Networks (DRAGON) campaign data obtained in the summer of 2011. These data, which were not available in a previous study, contain spatially dense ground measurements of the AOD and other aerosol particle characteristics from the Baltimore-Washington, DC region. Our results show that the HB-MCMC algorithm has 16.2% more AOD retrieval coverage and improves the root-mean-square error by 38.3% compared with the MISR operational algorithm. Our detailed analyses with various metrics show that the improvement of our scheme is coming from the novel modeling and inference method. Furthermore, the map overlay of the retrieval results qualitatively confirms the findings of the quantitative analyses.

Web portal services have become an important medium to deliver digital content (e.g. news, advertisements, etc.) to Web users in a timely fashion. To attract more users to various content modules on the Web portal, it is necessary to design a recommender system that can effectively achieve Web portal content optimization by automatically estimating content item attractiveness and relevance to user interests. The state-of-the-art online learning methodology adapts dedicated pointwise models to independently estimate the attractiveness score for each candidate content item. Although such pointwise models can be easily adapted for online recommendation, there still remain a few critical problems. First, this pointwise methodology fails to use invaluable user preferences between content items. Moreover, the performance of pointwise models decreases drastically when facing the problem of sparse learning samples. To address these problems, we propose exploring a new dynamic pairwise learning methodology for Web portal content optimization in which we exploit dynamic user preferences extracted based on users' actions on portal services to compute the attractiveness scores of content items. In this article, we introduce two specific pairwise learning algorithms, a straightforward graph-based algorithm and a formalized Bayesian modeling one. Experiments on large-scale data from a commercial Web portal demonstrate the significant improvement of pairwise methodologies over the baseline pointwise models. Further analysis illustrates that our new pairwise learning approaches can benefit personalized recommendation more than pointwise models, since the data sparsity is more critical for personalized content optimization.

Traditional machine-learned ranking systems for Web search are often trained to capture stationary relevance of documents to queries, which have limited ability to track nonstationary user intention in a timely manner. In recency search, for instance, the relevance of documents to a query on breaking news often changes significantly over time, requiring effective adaptation to user intention. In this article, we focus on recency search and study a number of algorithms to improve ranking results by leveraging user click feedback. Our contributions are threefold. First, we use commercial search engine sessions collected in a random exploration bucket for reliable offline evaluation of these algorithms, which provides an unbiased comparison across algorithms without online bucket tests. Second, we propose an online learning approach that reranks and improves the search results for recency queries near real-time based on user clicks. This approach is very general and can be combined with sophisticated click models. Third, our empirical comparison of a dozen algorithms on real-world search data suggests importance of a few algorithmic choices in these applications, including generalization across different query-document pairs, specialization to popular queries, and near real-time adaptation of user clicks for reranking.

We consider discrete denoising of two-dimensional data with characteristics that may be varying abruptly between regions. Using a quadtree decomposition technique and space-filling curves, we extend the recently developed S-DUDE (Shifting Discrete Universal DEnoiser), which was tailored to one-dimensional data, to the two-dimensional case. Our scheme competes with a genie that has access, in addition to the noisy data, also to the underlying noiseless data, and can employ m different two-dimensional sliding window denoisers along m distinct regions obtained by a quadtree decomposition with m leaves, in a way that minimizes the overall loss. We show that, regardless of what the underlying noiseless data may be, the two-dimensional S-DUDE performs essentially as well as this genie, provided that the number of distinct regions satisfies m = o(n), where n is the total size of the data. The resulting algorithm complexity is still linear in both n and m, as in the one-dimensional case. Our experimental results show that the two-dimensional S-DUDE can be effective when the characteristics of the underlying clean image vary across different regions in the data.

With the explosive growth of online news readership, recommending interesting news articles to users has become extremely important. While existing Web services such as Yahoo! and Digg attract users' initial clicks by leveraging various kinds of signals, how to engage such users algorithmically after their initial visit is largely under-explored. In this paper, we study the problem of post-click news recommendation. Given that a user has perused a current news article, our idea is to automatically identify "related" news articles which the user would like to read afterwards. Specifically, we propose to characterize relatedness between news articles across four aspects: relevance, novelty, connection clarity, and transition smoothness. Motivated by this understanding, we define a set of features to capture each of these aspects and put forward a learning approach to model relatedness. In order to quantitatively evaluate our proposed measures and learn a unified relatedness function, we construct a large test collection based on a four-month commercial news corpus with editorial judgments. The experimental results show that the proposed heuristics can indeed capture relatedness, and that the learned unified relatedness function works quite effectively.

We explore the potential of using users click-through logs where no editorial judgment is available to improve the ranking function of a vertical search engine. We base our analysis on the Cumulate Relevance Model, a user behavior model recently proposed as a way to extract relevance signal from click-through logs. We propose a novel way of directly learning the ranking function, effectively by-passing the need to have explicit editorial relevance label for each query-document pair. This approach potentially adjusts more closely the ranking function to a variety of user behaviors both at the individual and at the aggregate levels. We investigate two ways of using behavioral model; First, we consider the parametric approach where we learn the estimates of document relevance and use them as targets for the machine learned ranking schemes. In the second, functional approach, we learn a function that maximizes the behavioral model likelihood, effectively by-passing the need to estimate a substitute for document labels. Experiments using user session data collected from a commercial vertical search engine demonstrate the potential of our approach. While in terms of DCG, the editorial model out-perform the behavioral one, online experiments show that the behavioral model is on par --if not superior-- to the editorial model. To our knowledge, this is the first report in the Literature of a competitive behavioral model in a commercial setting

Traditional machine-learned ranking algorithms for web search are trained in batch mode, which assume static relevance of documents for a given query. Although such a batch-learning framework has been tremendously successful in commercial search engines, in scenarios where relevance of documents to a query changes over time, such as ranking recent documents for a breaking news query, the batch-learned ranking functions do have limitations. Users' real-time click feedback becomes a better and timely proxy for the varying relevance of documents rather than the editorial judgments provided by human editors. In this paper, we propose an online learning algorithm that can quickly learn the best re-ranking of the top portion of the original ranked list based on real-time users' click feedback. In order to devise our algorithm and evaluate it accurately, we collected exploration bucket data that removes positional biases on clicks on the documents for recency-classified queries. Our initial experimental result shows that our scheme is more capable of quickly adjusting the ranking to track the varying relevance of documents reflected in the click feedback, compared to batch-trained ranking functions.

Ranking a set of retrieved documents according to their relevance to a given query has become a popular problem at the intersection of web search, machine learning, and information retrieval. Recent work on ranking focused on a number of different paradigms, namely, pointwise, pairwise, and list-wise approaches. Each of those paradigms focuses on a different aspect of the dataset while largely ignoring others. The current paper shows how a combination of them can lead to improved ranking performance and, moreover, how it can be implemented in log-linear time.

We introduce S-DUDE, a new algorithm for denoising discrete memoryless channel (DMC)-corrupted data. The algorithm, which generalizes the recently introduced DUDE (discrete universal denoiser), aims to compete with a genie that has access, in addition to the noisy data, also to the underlying clean data, and that can choose to switch, up tomtimes, between sliding-window denoisers in a way that minimizes the overall loss. When the underlying data form an individual sequence, we show that the S-DUDE performs essentially as well as this genie, provided thatmis sublinear in the size of the data. When the clean data are emitted by a piecewise stationary process, we show that the S-DUDE achieves the optimum distribution-dependent performance, provided that the same sublinearity condition is imposed on the number of switches. To further substantiate the universal optimality of the S-DUDE, we show that when the number of switches is allowed to grow linearly with the size of the data,any(sequence of) scheme(s) fails to compete in the above sense. Using dynamic programming, we derive an efficient implementation of the S-DUDE, which has complexity (time and memory) growing linearly with the data size and the number of switchesm. Preliminary experimental results are presented, suggesting that S-DUDE has the capacity to improve on the performance attained by the original DUDE in applications where the nature of the data abruptly changes in time (or space), as is often the case in practice.

The problem of discrete universal filtering, in which the components of a discrete signal emitted by an unknown source and corrupted by a known discrete memoryless channel (DMC) are to be causally estimated, is considered. A family of filters are derived, and are shown to be universally asymptotically optimal in the sense of achieving the optimum filtering performance when the clean signal is stationary, ergodic, and satisfies an additional mild positivity condition. Our schemes are comprised of approximating the noisy signal using a hidden Markov process (HMP) via maximum-likelihood (ML) estimation, followed by the use of the forward recursions for HMP state estimation. It is shown that as the data length increases, and as the number of states in the HMP approximation increases, our family of filters attains the performance of the optimal distribution-dependent filter. An extension to the case of channels with memory is also established.

We consider the problem of causal estimation, i.e., filtering, of a real-valued signal corrupted by zero mean, i.i.d., real-valued additive noise under the mean square error (MSE) criterion. We build a competitive on-line filtering algorithm whose normalized cumulative MSE, for every bounded underlying signal, is asymptotically as small as the best linear finite-duration impulse response (FIR) filter of order d. We do not assume any stochastic mechanism in generating the underlying signal, and assume only the variance of the noise is known to the filter. The regret of our scheme is shown to decay in the order of O (log n/n), where n is the length of the signal. Moreover, we present a concentration of the average square error of our scheme to that of the best d-th order linear FIR filter. Our analysis combines tools from the problems of universal filtering and competitive on-line regression.

We consider the discrete universal filtering problem, where the components of a discrete signal emitted by an unknown source and corrupted by a known DMC are to be causally estimated. We derive a family of filters which we show to be universally asymptotically optimal in the sense of achieving the optimum filtering performance when the clean signal is stationary, ergodic, and satisfies an additional mild positivity condition. Our schemes are based on approximating the noisy signal by a hidden Markov process (HMP) via maximum likelihood (ML) estimation, followed by use of the well-known forward recursions for HMP state estimation. We show that as the data length increases, and as the number of states in the HMP approximation increases, our family of filters attain the performance of the optimal distribution-dependent filter