Adaptive Moment Estimation (Adam), which combines Adaptive Learning Rate and Momentum, would be the most popular stochastic optimizer for accelerating the training of deep neural networks. However, it is empirically known that Adam often generalizes worse than Stochastic Gradient Descent (SGD). The purpose of this paper is to unveil the mystery of this behavior in the diffusion theoretical framework. Specifically, we disentangle the effects of Adaptive Learning Rate and Momentum of the Adam dynamics on saddle-point escaping and flat minima selection. We prove that Adaptive Learning Rate can escape saddle points efficiently, but cannot select flat minima as SGD does. In contrast, Momentum provides a drift effect to help the training process pass through saddle points, and almost does not affect flat minima selection. This partly explains why SGD (with Momentum) generalizes better, while Adam generalizes worse but converges faster. Furthermore, motivated by the analysis, we design a novel adaptive optimization framework named Adaptive Inertia, which uses parameter-wise adaptive inertia to accelerate the training and provably favors flat minima as well as SGD. Our extensive experiments demonstrate that the proposed adaptive inertia method can generalize significantly better than SGD and conventional adaptive gradient methods.

Quantized neural network training optimizes a discrete, non-differentiable objective. The straight-through estimator (STE) enables backpropagation through surrogate gradients and is widely used. While previous studies have primarily focused on the properties of surrogate gradients and their convergence, the influence of quantization hyperparameters, such as bit width and quantization range, on learning dynamics remains largely unexplored. We theoretically show that in the high-dimensional limit, STE dynamics converge to a deterministic ordinary differential equation. This reveals that STE training exhibits a plateau followed by a sharp drop in generalization error, with plateau length depending on the quantization range. A fixed-point analysis quantifies the asymptotic deviation from the unquantized linear model. We also extend analytical techniques for stochastic gradient descent to nonlinear transformations of weights and inputs.

Weight decay is a simple yet powerful regularization technique that has been very widely used in training of deep neural networks (DNNs). While weight decay has attracted much attention, previous studies fail to discover some overlooked pitfalls on large gradient norms resulted by weight decay. In this paper, we discover that, weight decay can unfortunately lead to large gradient norms at the final phase (or the terminated solution) of training, which often indicates bad convergence and poor generalization. To mitigate the gradient-norm-centered pitfalls, we present the first practical scheduler for weight decay, called the Scheduled Weight Decay (SWD) method that can dynamically adjust the weight decay strength according to the gradient norm and significantly penalize large gradient norms during training. Our experiments also support that SWD indeed mitigates large gradient norms and often significantly outperforms the conventional constant weight decay strategy for Adaptive Moment Estimation (Adam).

People usually believe that network pruning not only reduces the computational cost of deep networks, but also prevents overfitting by decreasing model capacity. However, our work surprisingly discovers that network pruning sometimes even aggravates overfitting. We report an unexpected sparse double descent phenomenon that, as we increase model sparsity via network pruning, test performance first gets worse (due to overfitting), then gets better (due to relieved overfitting), and gets worse at last (due to forgetting useful information). While recent studies focused on the deep double descent with respect to model overparameterization, they failed to recognize that sparsity may also cause double descent. In this paper, we have three main contributions. First, we report the novel sparse double descent phenomenon through extensive experiments. Second, for this phenomenon, we propose a novel learning distance interpretation that the curve of $\ell_{2}$ learning distance of sparse models (from initialized parameters to final parameters) may correlate with the sparse double descent curve well and reflect generalization better than minima flatness. Third, in the context of sparse double descent, a winning ticket in the lottery ticket hypothesis surprisingly may not always win.

Open-vocabulary object detection (OVD), detecting specific classes of objects using only their linguistic descriptions (e.g., class names) without any image samples, has garnered significant attention. However, in real-world applications, the target class concepts is often hard to describe in text and the only way to specify target objects is to provide their image examples, yet it is often challenging to obtain a good number of samples. Thus, there is a high demand from practitioners for few-shot object detection (FSOD). A natural question arises: Can the benefits of OVD extend to FSOD for object classes that are difficult to describe in text? Compared to traditional methods that learn only predefined classes (referred to in this paper as closed-set object detection, COD), can the extra cost of OVD be justified? To answer these questions, we propose a method to quantify the ``text-describability'' of object detection datasets using the zero-shot image classification accuracy with CLIP. This allows us to categorize various OD datasets with different text-describability and emprically evaluate the FSOD performance of OVD and COD methods within each category. Our findings reveal that: i) there is little difference between OVD and COD for object classes with low text-describability under equal conditions in OD pretraining; and ii) although OVD can learn from more diverse data than OD-specific data, thereby increasing the volume of training data, it can be counterproductive for classes with low-text-describability. These findings provide practitioners with valuable guidance amidst the recent advancements of OVD methods.

Detecting objects occupying only small areas in an image is difficult, even for humans. Therefore, annotating small-size object instances is hard and thus costly. This study questions common sense by asking the following: is annotating small-size instances worth its cost? We restate it as the following verifiable question: can we detect small-size instances with a detector trained using training data free of small-size instances? We evaluate a method that upscales input images at test time and a method that downscales images at training time. The experiments conducted using the COCO dataset show the following. The first method, together with a remedy to narrow the domain gap between training and test inputs, achieves at least comparable performance to the baseline detector trained using complete training data. Although the method needs to apply the same detector twice to an input image with different scaling, we show that its distillation yields a single-path detector that performs equally well to the same baseline detector. These results point to the necessity of rethinking the annotation of training data for object detection.

In this paper, we propose a fast method for exactly enumerating a very large number of all lower cost solutions for various combinatorial problems. Our method is based on backtracking for a given decision diagram which represents all the feasible solutions. The main idea is to memoize the intervals of cost bounds to avoid duplicate search in the backtracking process. In contrast to usual pseudo-polynomial-time dynamic programming approaches, the computation time of our method does not directly depend on the total cost values, but is bounded by the input and output size of the decision diagrams. Therefore, it can be much faster if the cost values are large but the input/output decision diagrams are well-compressed. We demonstrate its practical efficiency by comparing our method to current available enumeration methods: for nontrivial size instances of the Hamiltonian path problem, our method succeeded in exactly enumerating billions of all lower cost solutions in a few seconds, which was hundred or much more times faster. Our method can be regarded as a novel search algorithm which integrates the two classical techniques, branch-and-bound and dynamic programming. This method would have many applications in various fields, including operations research, data mining, statistical testing, hardware/software system design, etc.

Smartphones with multi-camera systems, featuring cameras with varying field-of-views (FoVs), are increasingly common. This variation in FoVs results in content differences across videos, paving the way for an innovative approach to video super-resolution (VSR). This method enhances the VSR performance of lower resolution (LR) videos by leveraging higher resolution reference (Ref) videos. Previous works, which operate on this principle, generally expand on traditional VSR models by combining LR and Ref inputs over time into a unified stream. However, we can expect that better results are obtained by independently aggregating these Ref image sequences temporally. Therefore, we introduce an improved method, RefVSR++, which performs the parallel aggregation of LR and Ref images in the temporal direction, aiming to optimize the use of the available data. RefVSR++ also incorporates improved mechanisms for aligning image features over time, crucial for effective VSR. Our experiments demonstrate that RefVSR++ outperforms previous works by over 1dB in PSNR, setting a new benchmark in the field.

African languages have far less in-language content available digitally, making it challenging for question answering systems to satisfy the information needs of users. Cross-lingual open-retrieval question answering (XOR QA) systems -- those that retrieve answer content from other languages while serving people in their native language -- offer a means of filling this gap. To this end, we create AfriQA, the first cross-lingual QA dataset with a focus on African languages. AfriQA includes 12,000+ XOR QA examples across 10 African languages. While previous datasets have focused primarily on languages where cross-lingual QA augments coverage from the target language, AfriQA focuses on languages where cross-lingual answer content is the only high-coverage source of answer content. Because of this, we argue that African languages are one of the most important and realistic use cases for XOR QA. Our experiments demonstrate the poor performance of automatic translation and multilingual retrieval methods. Overall, AfriQA proves challenging for state-of-the-art QA models. We hope that the dataset enables the development of more equitable QA technology.

Diffusion models have demonstrated impressive image generation capabilities. Personalized approaches, such as textual inversion and Dreambooth, enhance model individualization using specific images. These methods enable generating images of specific objects based on diverse textual contexts. Our proposed approach aims to retain the model's original knowledge during new information integration, resulting in superior outcomes while necessitating less training time compared to Dreambooth and textual inversion.

In this paper, a bridge member damage cause estimation framework is proposed by calculating the image position using Structure from Motion (SfM) and acquiring its information via Visual Question Answering (VQA). For this, a VQA model was developed that uses bridge images for dataset creation and outputs the damage or member name and its existence based on the images and questions. In the developed model, the correct answer rate for questions requiring the member's name and the damage's name were 67.4% and 68.9%, respectively. The correct answer rate for questions requiring a yes/no answer was 99.1%. Based on the developed model, a damage cause estimation method was proposed. In the proposed method, the damage causes are narrowed down by inputting new questions to the VQA model, which are determined based on the surrounding images obtained via SfM and the results of the VQA model. Subsequently, the proposed method was then applied to an actual bridge and shown to be capable of determining damage and estimating its cause. The proposed method could be used to prevent damage causes from being overlooked, and practitioners could determine inspection focus areas, which could contribute to the improvement of maintenance techniques. In the future, it is expected to contribute to infrastructure diagnosis automation.

Advanced visual localization techniques encompass image retrieval challenges and 6 Degree-of-Freedom (DoF) camera pose estimation, such as hierarchical localization. Thus, they must extract global and local features from input images. Previous methods have achieved this through resource-intensive or accuracy-reducing means, such as combinatorial pipelines or multi-task distillation. In this study, we present a novel method called SuperGF, which effectively unifies local and global features for visual localization, leading to a higher trade-off between localization accuracy and computational efficiency. Specifically, SuperGF is a transformer-based aggregation model that operates directly on image-matching-specific local features and generates global features for retrieval. We conduct experimental evaluations of our method in terms of both accuracy and efficiency, demonstrating its advantages over other methods. We also provide implementations of SuperGF using various types of local features, including dense and sparse learning-based or hand-crafted descriptors.

In this research, we investigate the high-dimensional linear contextual bandit problem where the number of features $p$ is greater than the budget $T$, or it may even be infinite. Differing from the majority of previous works in this field, we do not impose sparsity on the regression coefficients. Instead, we rely on recent findings on overparameterized models, which enables us to analyze the performance the minimum-norm interpolating estimator when data distributions have small effective ranks. We propose an explore-then-commit (EtC) algorithm to address this problem and examine its performance. Through our analysis, we derive the optimal rate of the ETC algorithm in terms of $T$ and show that this rate can be achieved by balancing exploration and exploitation. Moreover, we introduce an adaptive explore-then-commit (AEtC) algorithm that adaptively finds the optimal balance. We assess the performance of the proposed algorithms through a series of simulations.

Investigating critical phenomena or phase transitions is of high interest in physics and chemistry, for which Monte Carlo (MC) simulations, a crucial tool for numerically analyzing macroscopic properties of given systems, are often hindered by an emerging divergence of correlation length -- known as scale invariance at criticality (SIC) in the renormalization group theory. SIC causes the system to behave the same at any length scale, from which many existing sampling methods suffer: long-range correlations cause critical slowing down in Markov chain Monte Carlo (MCMC), and require intractably large receptive fields for generative samplers. In this paper, we propose a Renormalization-informed Generative Critical Sampler (RiGCS) -- a novel sampler specialized for near-critical systems, where SIC is leveraged as an advantage rather than a nuisance. Specifically, RiGCS builds on MultiLevel Monte Carlo (MLMC) with Heat Bath (HB) algorithms, which perform ancestral sampling from low-resolution to high-resolution lattice configurations with site-wise-independent conditional HB sampling. Although MLMC-HB is highly efficient under exact SIC, it suffers from a low acceptance rate under slight SIC violation. Notably, SIC violation always occurs in finite-size systems, and may induce long-range and higher-order interactions in the renormalized distributions, which are not considered by independent HB samplers. RiGCS enhances MLMC-HB by replacing a part of the conditional HB sampler with generative models that capture those residual interactions and improve the sampling efficiency. Our experiments show that the effective sample size of RiGCS is a few orders of magnitude higher than state-of-the-art generative model baselines in sampling configurations for 128x128 two-dimensional Ising systems.

We introduce and study a new variant of the multi-armed bandit problem (MAB), called the survival bandit problem (S-MAB). While in both problems, the objective is to maximize the so-called cumulative reward, in this new variant, the procedure is interrupted if the cumulative reward falls below a preset threshold. This simple yet unexplored extension of the MAB follows from many practical applications. For example, when testing two medicines against each other on voluntary patients, people's health are at stake, and it is necessary to be able to interrupt experiments if serious side effects occur or if the disease syndromes are not dissipated by the treatment. From a theoretical perspective, the S-MAB is the first variant of the MAB where the procedure may or may not be interrupted. We start by formalizing the S-MAB and we define its objective as the minimization of the so-called survival regret, which naturally generalizes the regret of the MAB. Then, we show that the objective of the S-MAB is considerably more difficult than the MAB, in the sense that contrary to the MAB, no policy can achieve a reasonably small (i.e., sublinear) survival regret. Instead, we minimize the survival regret in the sense of Pareto, i.e., we seek a policy whose cumulative reward cannot be improved for some problem instance without being sacrificed for another one. For that purpose, we identify two key components in the survival regret: the regret given no ruin (which corresponds to the regret in the MAB), and the probability that the procedure is interrupted, called the probability of ruin. We derive a lower bound on the probability of ruin, as well as policies whose probability of ruin matches the lower bound. Finally, based on a doubling trick on those policies, we derive a policy which minimizes the survival regret in the sense of Pareto, giving an answer to an open problem by Perotto et al. (COLT 2019).

It is well-known that stochastic gradient noise (SGN) acts as implicit regularization for deep learning and is essentially important for both optimization and generalization of deep networks. Some works attempted to artificially simulate SGN by injecting random noise to improve deep learning. However, it turned out that the injected simple random noise cannot work as well as SGN, which is anisotropic and parameter-dependent. For simulating SGN at low computational costs and without changing the learning rate or batch size, we propose the Positive-Negative Momentum (PNM) approach that is a powerful alternative to conventional Momentum in classic optimizers. The introduced PNM method maintains two approximate independent momentum terms. Then, we can control the magnitude of SGN explicitly by adjusting the momentum difference. We theoretically prove the convergence guarantee and the generalization advantage of PNM over Stochastic Gradient Descent (SGD). By incorporating PNM into the two conventional optimizers, SGD with Momentum and Adam, our extensive experiments empirically verified the significant advantage of the PNM-based variants over the corresponding conventional Momentum-based optimizers.

We propose an algorithm to estimate the path-gradient of both the reverse and forward Kullback-Leibler divergence for an arbitrary manifestly invertible normalizing flow. The resulting path-gradient estimators are straightforward to implement, have lower variance, and lead not only to faster convergence of training but also to better overall approximation results compared to standard total gradient estimators. We also demonstrate that path-gradient training is less susceptible to mode-collapse. In light of our results, we expect that path-gradient estimators will become the new standard method to train normalizing flows for variational inference.