Search

The integration of vision-language models such as CLIP and Concept Bottleneck Models (CBMs) offers a promising approach to explaining deep neural network (DNN) decisions using concepts understandable by humans, addressing the black-box concern of DNNs. While CLIP provides both explainability and zero-shot classification capability, its pre-training on generic image and text data may limit its classification accuracy and applicability to medical image diagnostic tasks, creating a transfer learning problem. To maintain explainability and address transfer learning needs, CBM methods commonly design post-processing modules after the bottleneck module. However, this way has been ineffective. This paper takes an unconventional approach by re-examining the CBM framework through the lens of its geometrical representation as a simple linear classification system. The analysis uncovers that post-CBM fine-tuning modules merely rescale and shift the classification outcome of the system, failing to fully leverage the system's learning potential. We introduce an adaptive module strategically positioned between CLIP and CBM to bridge the gap between source and downstream domains. This simple yet effective approach enhances classification performance while preserving the explainability afforded by the framework. Our work offers a comprehensive solution that encompasses the entire process, from concept discovery to model training, providing a holistic recipe for leveraging the strengths of GPT, CLIP, and CBM., Comment: Accepted at MICCAI 2024, the 27th International Conference on Medical Image Computing and Computer Assisted Intervention

Automatically generating symbolic music-music scores tailored to specific human needs-can be highly beneficial for musicians and enthusiasts. Recent studies have shown promising results using extensive datasets and advanced transformer architectures. However, these state-of-the-art models generally offer only basic control over aspects like tempo and style for the entire composition, lacking the ability to manage finer details, such as control at the level of individual bars. While fine-tuning a pre-trained symbolic music generation model might seem like a straightforward method for achieving this finer control, our research indicates challenges in this approach. The model often fails to respond adequately to new, fine-grained bar-level control signals. To address this, we propose two innovative solutions. First, we introduce a pre-training task designed to link control signals directly with corresponding musical tokens, which helps in achieving a more effective initialization for subsequent fine-tuning. Second, we implement a novel counterfactual loss that promotes better alignment between the generated music and the control prompts. Together, these techniques significantly enhance our ability to control music generation at the bar level, showing a 13.06\% improvement over conventional methods. Our subjective evaluations also confirm that this enhanced control does not compromise the musical quality of the original pre-trained generative model., Comment: Demo is available at: https://ganperf.github.io/musebarcontrol.github.io/musebarcontrol/

Pre-trained models produce strong generic representations that can be adapted via fine-tuning. The learned weight difference relative to the pre-trained model, known as a task vector, characterises the direction and stride of fine-tuning. The significance of task vectors is such that simple arithmetic operations on them can be used to combine diverse representations from different domains. This paper builds on these properties of task vectors and aims to answer (1) whether components of task vectors, particularly parameter blocks, exhibit similar characteristics, and (2) how such blocks can be used to enhance knowledge composition and transfer. To this end, we introduce aTLAS, an algorithm that linearly combines parameter blocks with different learned coefficients, resulting in anisotropic scaling at the task vector level. We show that such linear combinations explicitly exploit the low intrinsic dimensionality of pre-trained models, with only a few coefficients being the learnable parameters. Furthermore, composition of parameter blocks leverages the already learned representations, thereby reducing the dependency on large amounts of data. We demonstrate the effectiveness of our method in task arithmetic, few-shot recognition and test-time adaptation, with supervised or unsupervised objectives. In particular, we show that (1) learned anisotropic scaling allows task vectors to be more disentangled, causing less interference in composition; (2) task vector composition excels with scarce or no labeled data and is less prone to domain shift, thus leading to better generalisability; (3) mixing the most informative parameter blocks across different task vectors prior to training can reduce the memory footprint and improve the flexibility of knowledge transfer. Moreover, we show the potential of aTLAS as a PEFT method, particularly with less data, and demonstrate that its scalibility.

Contrastive Language-Image Pretraining (CLIP) stands out as a prominent method for image representation learning. Various architectures, from vision transformers (ViTs) to convolutional networks (ResNets) have been trained with CLIP to serve as general solutions to diverse vision tasks. This paper explores the differences across various CLIP-trained vision backbones. Despite using the same data and training objective, we find that these architectures have notably different representations, different classification performance across datasets, and different robustness properties to certain types of image perturbations. Our findings indicate a remarkable possible synergy across backbones by leveraging their respective strengths. In principle, classification accuracy could be improved by over 40 percentage with an informed selection of the optimal backbone per test example.Using this insight, we develop a straightforward yet powerful approach to adaptively ensemble multiple backbones. The approach uses as few as one labeled example per class to tune the adaptive combination of backbones. On a large collection of datasets, the method achieves a remarkable increase in accuracy of up to 39.1% over the best single backbone, well beyond traditional ensembles, Comment: arXiv admin note: substantial text overlap with arXiv:2312.14400

For many recommender systems the primary data source is a historical record of user clicks. The associated click matrix which is often very sparse, however, as the number of users x products can be far larger than the number of clicks, and such sparsity is accentuated in cold-start settings. The sparsity of the click matrix is the reason matrix factorization and autoencoders techniques remain highly competitive across collaborative filtering datasets. In this work, we propose a simple approach to address cold-start recommendations by leveraging content metadata, Metadata Alignment for cold-start Recommendation. we show that this approach can readily augment existing matrix factorization and autoencoder approaches, enabling a smooth transition to top performing algorithms in warmer set-ups. Our experimental results indicate three separate contributions: first, we show that our proposed framework largely beats SOTA results on 4 cold-start datasets with different sparsity and scale characteristics, with gains ranging from +8.4% to +53.8% on reported ranking metrics; second, we provide an ablation study on the utility of semantic features, and proves the additional gain obtained by leveraging such features ranges between +46.8% and +105.5%; and third, our approach is by construction highly competitive in warm set-ups, and we propose a closed-form solution outperformed by SOTA results by only 0.8% on average.

Deep Neural Networks (DNNs) are widely used for visual classification tasks, but their complex computation process and black-box nature hinder decision transparency and interpretability. Class activation maps (CAMs) and recent variants provide ways to visually explain the DNN decision-making process by displaying 'attention' heatmaps of the DNNs. Nevertheless, the CAM explanation only offers relative attention information, that is, on an attention heatmap, we can interpret which image region is more or less important than the others. However, these regions cannot be meaningfully compared across classes, and the contribution of each region to the model's class prediction is not revealed. To address these challenges that ultimately lead to better DNN Interpretation, in this paper, we propose CAPE, a novel reformulation of CAM that provides a unified and probabilistically meaningful assessment of the contributions of image regions. We quantitatively and qualitatively compare CAPE with state-of-the-art CAM methods on CUB and ImageNet benchmark datasets to demonstrate enhanced interpretability. We also test on a cytology imaging dataset depicting a challenging Chronic Myelomonocytic Leukemia (CMML) diagnosis problem. Code is available at: https://github.com/AIML-MED/CAPE.

Causal representation learning seeks to uncover latent, high-level causal representations from low-level observed data. It is particularly good at predictions under unseen distribution shifts, because these shifts can generally be interpreted as consequences of interventions. Hence leveraging {seen} distribution shifts becomes a natural strategy to help identifying causal representations, which in turn benefits predictions where distributions are previously {unseen}. Determining the types (or conditions) of such distribution shifts that do contribute to the identifiability of causal representations is critical. This work establishes a {sufficient} and {necessary} condition characterizing the types of distribution shifts for identifiability in the context of latent additive noise models. Furthermore, we present partial identifiability results when only a portion of distribution shifts meets the condition. In addition, we extend our findings to latent post-nonlinear causal models. We translate our findings into a practical algorithm, allowing for the acquisition of reliable latent causal representations. Our algorithm, guided by our underlying theory, has demonstrated outstanding performance across a diverse range of synthetic and real-world datasets. The empirical observations align closely with the theoretical findings, affirming the robustness and effectiveness of our approach.

Continual learning requires a model to adapt to ongoing changes in the data distribution, and often to the set of tasks to be performed. It is rare, however, that the data and task changes are completely unpredictable. Given a description of an overarching goal or data theme, which we call a realm, humans can often guess what concepts are associated with it. We show here that the combination of a large language model and an image generation model can similarly provide useful premonitions as to how a continual learning challenge might develop over time. We use the large language model to generate text descriptions of semantically related classes that might potentially appear in the data stream in future. These descriptions are then rendered using Stable Diffusion to generate new labelled image samples. The resulting synthetic dataset is employed for supervised pre-training, but is discarded prior to commencing continual learning, along with the pre-training classification head. We find that the backbone of our pre-trained networks can learn representations useful for the downstream continual learning problem, thus becoming a valuable input to any existing continual learning method. Although there are complexities arising from the domain gap between real and synthetic images, we show that pre-training models in this manner improves multiple Class Incremenal Learning (CIL) methods on fine-grained image classification benchmarks. Supporting code can be found at https://github.com/cl-premonition/premonition., Comment: 31 pages total (14 main paper, 5 references, 12 appendices)

Novel-view synthesis through diffusion models has demonstrated remarkable potential for generating diverse and high-quality images. Yet, the independent process of image generation in these prevailing methods leads to challenges in maintaining multiple-view consistency. To address this, we introduce ViewFusion, a novel, training-free algorithm that can be seamlessly integrated into existing pre-trained diffusion models. Our approach adopts an auto-regressive method that implicitly leverages previously generated views as context for the next view generation, ensuring robust multi-view consistency during the novel-view generation process. Through a diffusion process that fuses known-view information via interpolated denoising, our framework successfully extends single-view conditioned models to work in multiple-view conditional settings without any additional fine-tuning. Extensive experimental results demonstrate the effectiveness of ViewFusion in generating consistent and detailed novel views., Comment: CVPR2024,homepage:https://wi-sc.github.io/ViewFusion.github.io/

Given a script, the challenge in Movie Dubbing (Visual Voice Cloning, V2C) is to generate speech that aligns well with the video in both time and emotion, based on the tone of a reference audio track. Existing state-of-the-art V2C models break the phonemes in the script according to the divisions between video frames, which solves the temporal alignment problem but leads to incomplete phoneme pronunciation and poor identity stability. To address this problem, we propose StyleDubber, which switches dubbing learning from the frame level to phoneme level. It contains three main components: (1) A multimodal style adaptor operating at the phoneme level to learn pronunciation style from the reference audio, and generate intermediate representations informed by the facial emotion presented in the video; (2) An utterance-level style learning module, which guides both the mel-spectrogram decoding and the refining processes from the intermediate embeddings to improve the overall style expression; And (3) a phoneme-guided lip aligner to maintain lip sync. Extensive experiments on two of the primary benchmarks, V2C and Grid, demonstrate the favorable performance of the proposed method as compared to the current stateof-the-art. The code will be made available at https://github.com/GalaxyCong/StyleDubber.

Multimodal contrastive representation learning methods have proven successful across a range of domains, partly due to their ability to generate meaningful shared representations of complex phenomena. To enhance the depth of analysis and understanding of these acquired representations, we introduce a unified causal model specifically designed for multimodal data. By examining this model, we show that multimodal contrastive representation learning excels at identifying latent coupled variables within the proposed unified model, up to linear or permutation transformations resulting from different assumptions. Our findings illuminate the potential of pre-trained multimodal models, eg, CLIP, in learning disentangled representations through a surprisingly simple yet highly effective tool: linear independent component analysis. Experiments demonstrate the robustness of our findings, even when the assumptions are violated, and validate the effectiveness of the proposed method in learning disentangled representations.

Source-Free Unsupervised Domain Adaptation (SFUDA) is a challenging task where a model needs to be adapted to a new domain without access to target domain labels or source domain data. The primary difficulty in this task is that the model's predictions may be inaccurate, and using these inaccurate predictions for model adaptation can lead to misleading results. To address this issue, this paper proposes a novel approach that considers multiple prediction hypotheses for each sample and investigates the rationale behind each hypothesis. By consolidating these hypothesis rationales, we identify the most likely correct hypotheses, which we then use as a pseudo-labeled set to support a semi-supervised learning procedure for model adaptation. To achieve the optimal performance, we propose a three-step adaptation process: model pre-adaptation, hypothesis consolidation, and semi-supervised learning. Extensive experimental results demonstrate that our approach achieves state-of-the-art performance in the SFUDA task and can be easily integrated into existing approaches to improve their performance. The codes are available at \url{https://github.com/GANPerf/HCPR}.

Contrastive Language-Image Pretraining (CLIP) stands out as a prominent method for image representation learning. Various neural architectures, spanning Transformer-based models like Vision Transformers (ViTs) to Convolutional Networks (ConvNets) like ResNets, are trained with CLIP and serve as universal backbones across diverse vision tasks. Despite utilizing the same data and training objectives, the effectiveness of representations learned by these architectures raises a critical question. Our investigation explores the differences in CLIP performance among these backbone architectures, revealing significant disparities in their classifications. Notably, normalizing these representations results in substantial performance variations. Our findings showcase a remarkable possible synergy between backbone predictions that could reach an improvement of over 20% through informed selection of the appropriate backbone. Moreover, we propose a simple, yet effective approach to combine predictions from multiple backbones, leading to a notable performance boost of up to 6.34\%. We will release the code for reproducing the results.

Video Individual Counting (VIC) aims to predict the number of unique individuals in a single video. % Existing methods learn representations based on trajectory labels for individuals, which are annotation-expensive. % To provide a more realistic reflection of the underlying practical challenge, we introduce a weakly supervised VIC task, wherein trajectory labels are not provided. Instead, two types of labels are provided to indicate traffic entering the field of view (inflow) and leaving the field view (outflow). % We also propose the first solution as a baseline that formulates the task as a weakly supervised contrastive learning problem under group-level matching. In doing so, we devise an end-to-end trainable soft contrastive loss to drive the network to distinguish inflow, outflow, and the remaining. % To facilitate future study in this direction, we generate annotations from the existing VIC datasets SenseCrowd and CroHD and also build a new dataset, UAVVIC. % Extensive results show that our baseline weakly supervised method outperforms supervised methods, and thus, little information is lost in the transition to the more practically relevant weakly supervised task. The code and trained model will be public at \href{https://github.com/streamer-AP/CGNet}{CGNet}

Causal representation learning aims to unveil latent high-level causal representations from observed low-level data. One of its primary tasks is to provide reliable assurance of identifying these latent causal models, known as identifiability. A recent breakthrough explores identifiability by leveraging the change of causal influences among latent causal variables across multiple environments \citep{liu2022identifying}. However, this progress rests on the assumption that the causal relationships among latent causal variables adhere strictly to linear Gaussian models. In this paper, we extend the scope of latent causal models to involve nonlinear causal relationships, represented by polynomial models, and general noise distributions conforming to the exponential family. Additionally, we investigate the necessity of imposing changes on all causal parameters and present partial identifiability results when part of them remains unchanged. Further, we propose a novel empirical estimation method, grounded in our theoretical finding, that enables learning consistent latent causal representations. Our experimental results, obtained from both synthetic and real-world data, validate our theoretical contributions concerning identifiability and consistency.

This paper offers a new perspective to ease the challenge of domain generalization, which involves maintaining robust results even in unseen environments. Our design focuses on the decision-making process in the final classifier layer. Specifically, we propose treating the element-wise contributions to the final results as the rationale for making a decision and representing the rationale for each sample as a matrix. For a well-generalized model, we suggest the rationale matrices for samples belonging to the same category should be similar, indicating the model relies on domain-invariant clues to make decisions, thereby ensuring robust results. To implement this idea, we introduce a rationale invariance loss as a simple regularization technique, requiring only a few lines of code. Our experiments demonstrate that the proposed approach achieves competitive results across various datasets, despite its simplicity. Code is available at \url{https://github.com/liangchen527/RIDG}., Comment: Accepted in ICCV 2023

Continual learning (CL) aims to incrementally learn different tasks (such as classification) in a non-stationary data stream without forgetting old ones. Most CL works focus on tackling catastrophic forgetting under a learning-from-scratch paradigm. However, with the increasing prominence of foundation models, pre-trained models equipped with informative representations have become available for various downstream requirements. Several CL methods based on pre-trained models have been explored, either utilizing pre-extracted features directly (which makes bridging distribution gaps challenging) or incorporating adaptors (which may be subject to forgetting). In this paper, we propose a concise and effective approach for CL with pre-trained models. Given that forgetting occurs during parameter updating, we contemplate an alternative approach that exploits training-free random projectors and class-prototype accumulation, which thus bypasses the issue. Specifically, we inject a frozen Random Projection layer with nonlinear activation between the pre-trained model's feature representations and output head, which captures interactions between features with expanded dimensionality, providing enhanced linear separability for class-prototype-based CL. We also demonstrate the importance of decorrelating the class-prototypes to reduce the distribution disparity when using pre-trained representations. These techniques prove to be effective and circumvent the problem of forgetting for both class- and domain-incremental continual learning. Compared to previous methods applied to pre-trained ViT-B/16 models, we reduce final error rates by between 20% and 62% on seven class-incremental benchmarks, despite not using any rehearsal memory. We conclude that the full potential of pre-trained models for simple, effective, and fast CL has not hitherto been fully tapped. Code is at github.com/RanPAC/RanPAC., Comment: 32 pages, 11 figures

Gaussian process (GP) based Bayesian optimization (BO) is a powerful method for optimizing black-box functions efficiently. The practical performance and theoretical guarantees of this approach depend on having the correct GP hyperparameter values, which are usually unknown in advance and need to be estimated from the observed data. However, in practice, these estimations could be incorrect due to biased data sampling strategies used in BO. This can lead to degraded performance and break the sub-linear global convergence guarantee of BO. To address this issue, we propose a new BO method that can sub-linearly converge to the objective function's global optimum even when the true GP hyperparameters are unknown in advance and need to be estimated from the observed data. Our method uses a multi-armed bandit technique (EXP3) to add random data points to the BO process, and employs a novel training loss function for the GP hyperparameter estimation process that ensures consistent estimation. We further provide theoretical analysis of our proposed method. Finally, we demonstrate empirically that our method outperforms existing approaches on various synthetic and real-world problems., Comment: 25 pages, 5 figures

Rotated bounding boxes drastically reduce output ambiguity of elongated objects, making it superior to axis-aligned bounding boxes. Despite the effectiveness, rotated detectors are not widely employed. Annotating rotated bounding boxes is such a laborious process that they are not provided in many detection datasets where axis-aligned annotations are used instead. In this paper, we propose a framework that allows the model to predict precise rotated boxes only requiring cheaper axis-aligned annotation of the target dataset 1. To achieve this, we leverage the fact that neural networks are capable of learning richer representation of the target domain than what is utilized by the task. The under-utilized representation can be exploited to address a more detailed task. Our framework combines task knowledge of an out-of-domain source dataset with stronger annotation and domain knowledge of the target dataset with weaker annotation. A novel assignment process and projection loss are used to enable the co-training on the source and target datasets. As a result, the model is able to solve the more detailed task in the target domain, without additional computation overhead during inference. We extensively evaluate the method on various target datasets including fresh-produce dataset, HRSC2016 and SSDD. Results show that the proposed method consistently performs on par with the fully supervised approach., Comment: 10 pages, 5 figures, Accepted by CVPR 2023

Metric learning is a fundamental problem in computer vision whereby a model is trained to learn a semantically useful embedding space via ranking losses. Traditionally, the effectiveness of a ranking loss depends on the minibatch size, and is, therefore, inherently limited by the memory constraints of the underlying hardware. While simply accumulating the embeddings across minibatches has proved useful (Wang et al. [2020]), we show that it is equally important to ensure that the accumulated embeddings are up to date. In particular, it is necessary to circumvent the representational drift between the accumulated embeddings and the feature embeddings at the current training iteration as the learnable parameters are being updated. In this paper, we model representational drift as distribution misalignment and tackle it using moment matching. The result is a simple method for updating the stored embeddings to match the first and second moments of the current embeddings at each training iteration. Experiments on three popular image retrieval datasets, namely, SOP, In-Shop, and DeepFashion2, demonstrate that our approach significantly improves the performance in all scenarios.

Self-supervised learning (SSL) strategies have demonstrated remarkable performance in various recognition tasks. However, both our preliminary investigation and recent studies suggest that they may be less effective in learning representations for fine-grained visual recognition (FGVR) since many features helpful for optimizing SSL objectives are not suitable for characterizing the subtle differences in FGVR. To overcome this issue, we propose learning an additional screening mechanism to identify discriminative clues commonly seen across instances and classes, dubbed as common rationales in this paper. Intuitively, common rationales tend to correspond to the discriminative patterns from the key parts of foreground objects. We show that a common rationale detector can be learned by simply exploiting the GradCAM induced from the SSL objective without using any pre-trained object parts or saliency detectors, making it seamlessly to be integrated with the existing SSL process. Specifically, we fit the GradCAM with a branch with limited fitting capacity, which allows the branch to capture the common rationales and discard the less common discriminative patterns. At the test stage, the branch generates a set of spatial weights to selectively aggregate features representing an instance. Extensive experimental results on four visual tasks demonstrate that the proposed method can lead to a significant improvement in different evaluation settings., Comment: CVPR 2023

Reasoning the 3D structure of a non-rigid dynamic scene from a single moving camera is an under-constrained problem. Inspired by the remarkable progress of neural radiance fields (NeRFs) in photo-realistic novel view synthesis of static scenes, extensions have been proposed for dynamic settings. These methods heavily rely on neural priors in order to regularize the problem. In this work, we take a step back and reinvestigate how current implementations may entail deleterious effects, including limited expressiveness, entanglement of light and density fields, and sub-optimal motion localization. As a remedy, we advocate for a bridge between classic non-rigid-structure-from-motion (\nrsfm) and NeRF, enabling the well-studied priors of the former to constrain the latter. To this end, we propose a framework that factorizes time and space by formulating a scene as a composition of bandlimited, high-dimensional signals. We demonstrate compelling results across complex dynamic scenes that involve changes in lighting, texture and long-range dynamics.

The ability to use inductive reasoning to extract general rules from multiple observations is a vital indicator of intelligence. As humans, we use this ability to not only interpret the world around us, but also to predict the outcomes of the various interactions we experience. Generalising over multiple observations is a task that has historically presented difficulties for machines to grasp, especially when requiring computer vision. In this paper, we propose a model that can extract general rules from video demonstrations by simultaneously performing summarisation and translation. Our approach differs from prior works by framing the problem as a multi-sequence-to-sequence task, wherein summarisation is learnt by the model. This allows our model to utilise edge cases that would otherwise be suppressed or discarded by traditional summarisation techniques. Additionally, we show that our approach can handle noisy specifications without the need for additional filtering methods. We evaluate our model by synthesising programs from video demonstrations in the Vizdoom environment achieving state-of-the-art results with a relative increase of 11.75% program accuracy on prior works

Self-supervised learning (SSL) aims to produce useful feature representations without access to any human-labeled data annotations. Due to the success of recent SSL methods based on contrastive learning, such as SimCLR, this problem has gained popularity. Most current contrastive learning approaches append a parametrized projection head to the end of some backbone network to optimize the InfoNCE objective and then discard the learned projection head after training. This raises a fundamental question: Why is a learnable projection head required if we are to discard it after training? In this work, we first perform a systematic study on the behavior of SSL training focusing on the role of the projection head layers. By formulating the projection head as a parametric component for the InfoNCE objective rather than a part of the network, we present an alternative optimization scheme for training contrastive learning based SSL frameworks. Our experimental study on multiple image classification datasets demonstrates the effectiveness of the proposed approach over alternatives in the SSL literature.

Multi-source domain adaptation (MSDA) addresses the challenge of learning a label prediction function for an unlabeled target domain by leveraging both the labeled data from multiple source domains and the unlabeled data from the target domain. Conventional MSDA approaches often rely on covariate shift or conditional shift paradigms, which assume a consistent label distribution across domains. However, this assumption proves limiting in practical scenarios where label distributions do vary across domains, diminishing its applicability in real-world settings. For example, animals from different regions exhibit diverse characteristics due to varying diets and genetics. Motivated by this, we propose a novel paradigm called latent covariate shift (LCS), which introduces significantly greater variability and adaptability across domains. Notably, it provides a theoretical assurance for recovering the latent cause of the label variable, which we refer to as the latent content variable. Within this new paradigm, we present an intricate causal generative model by introducing latent noises across domains, along with a latent content variable and a latent style variable to achieve more nuanced rendering of observational data. We demonstrate that the latent content variable can be identified up to block identifiability due to its versatile yet distinct causal structure. We anchor our theoretical insights into a novel MSDA method, which learns the label distribution conditioned on the identifiable latent content variable, thereby accommodating more substantial distribution shifts. The proposed approach showcases exceptional performance and efficacy on both simulated and real-world datasets.

The task of causal representation learning aims to uncover latent higher-level causal representations that affect lower-level observations. Identifying true latent causal representations from observed data, while allowing instantaneous causal relations among latent variables, remains a challenge, however. To this end, we start from the analysis of three intrinsic properties in identifying latent space from observations: transitivity, permutation indeterminacy, and scaling indeterminacy. We find that transitivity acts as a key role in impeding the identifiability of latent causal representations. To address the unidentifiable issue due to transitivity, we introduce a novel identifiability condition where the underlying latent causal model satisfies a linear-Gaussian model, in which the causal coefficients and the distribution of Gaussian noise are modulated by an additional observed variable. Under some mild assumptions, we can show that the latent causal representations can be identified up to trivial permutation and scaling. Furthermore, based on this theoretical result, we propose a novel method, termed Structural caUsAl Variational autoEncoder, which directly learns latent causal representations and causal relationships among them, together with the mapping from the latent causal variables to the observed ones. We show that the proposed method learns the true parameters asymptotically. Experimental results on synthetic and real data demonstrate the identifiability and consistency results and the efficacy of the proposed method in learning latent causal representations.

Although Transformers have successfully transitioned from their language modelling origins to image-based applications, their quadratic computational complexity remains a challenge, particularly for dense prediction. In this paper we propose a content-based sparse attention method, as an alternative to dense self-attention, aiming to reduce the computation complexity while retaining the ability to model long-range dependencies. Specifically, we cluster and then aggregate key and value tokens, as a content-based method of reducing the total token count. The resulting clustered-token sequence retains the semantic diversity of the original signal, but can be processed at a lower computational cost. Besides, we further extend the clustering-guided attention from single-scale to multi-scale, which is conducive to dense prediction tasks. We label the proposed Transformer architecture ClusTR, and demonstrate that it achieves state-of-the-art performance on various vision tasks but at lower computational cost and with fewer parameters. For instance, our ClusTR small model with 22.7M parameters achieves 83.2\% Top-1 accuracy on ImageNet. Source code and ImageNet models will be made publicly available., Comment: 14 pages

The availability of clean and diverse labeled data is a major roadblock for training models on complex tasks such as visual question answering (VQA). The extensive work on large vision-and-language models has shown that self-supervised learning is effective for pretraining multimodal interactions. In this technical report, we focus on visual representations. We review and evaluate self-supervised methods to leverage unlabeled images and pretrain a model, which we then fine-tune on a custom VQA task that allows controlled evaluation and diagnosis. We compare energy-based models (EBMs) with contrastive learning (CL). While EBMs are growing in popularity, they lack an evaluation on downstream tasks. We find that both EBMs and CL can learn representations from unlabeled images that enable training a VQA model on very little annotated data. In a simple setting similar to CLEVR, we find that CL representations also improve systematic generalization, and even match the performance of representations from a larger, supervised, ImageNet-pretrained model. However, we find EBMs to be difficult to train because of instabilities and high variability in their results. Although EBMs prove useful for OOD detection, other results on supervised energy-based training and uncertainty calibration are largely negative. Overall, CL currently seems a preferable option over EBMs.

Semi-supervised learning is a critical tool in reducing machine learning's dependence on labeled data. It has been successfully applied to structured data, such as images and natural language, by exploiting the inherent spatial and semantic structure therein with pretrained models or data augmentation. These methods are not applicable, however, when the data does not have the appropriate structure, or invariances. Due to their simplicity, pseudo-labeling (PL) methods can be widely used without any domain assumptions. However, the greedy mechanism in PL is sensitive to a threshold and can perform poorly if wrong assignments are made due to overconfidence. This paper studies theoretically the role of uncertainty to pseudo-labeling and proposes Confident Sinkhorn Allocation (CSA), which identifies the best pseudo-label allocation via optimal transport to only samples with high confidence scores. CSA outperforms the current state-of-the-art in this practically important area of semi-supervised learning. Additionally, we propose to use the Integral Probability Metrics to extend and improve the existing PACBayes bound which relies on the Kullback-Leibler (KL) divergence, for ensemble models. Our code is publicly available at https://github.com/amzn/confident-sinkhorn-allocation., Comment: Code https://github.com/amzn/confident-sinkhorn-allocation

The current state-of-the-art methods in 3D instance segmentation typically involve a clustering step, despite the tendency towards heuristics, greedy algorithms, and a lack of robustness to the changes in data statistics. In contrast, we propose a fully-convolutional 3D point cloud instance segmentation method that works in a per-point prediction fashion. In doing so it avoids the challenges that clustering-based methods face: introducing dependencies among different tasks of the model. We find the key to its success is assigning a suitable target to each sampled point. Instead of the commonly used static or distance-based assignment strategies, we propose to use an Optimal Transport approach to optimally assign target masks to the sampled points according to the dynamic matching costs. Our approach achieves promising results on both ScanNet and S3DIS benchmarks. The proposed approach removes intertask dependencies and thus represents a simpler and more flexible 3D instance segmentation framework than other competing methods, while achieving improved segmentation accuracy., Comment: Accepted by ECCV22. Code and model will be released at https://github.com/tonghe90/PointInst3D

Convolutional neural networks (CNNs) have gained significant popularity in orthopedic imaging in recent years due to their ability to solve fracture classification problems. A common criticism of CNNs is their opaque learning and reasoning process, making it difficult to trust machine diagnosis and the subsequent adoption of such algorithms in clinical setting. This is especially true when the CNN is trained with limited amount of medical data, which is a common issue as curating sufficiently large amount of annotated medical imaging data is a long and costly process. While interest has been devoted to explaining CNN learnt knowledge by visualizing network attention, the utilization of the visualized attention to improve network learning has been rarely investigated. This paper explores the effectiveness of regularizing CNN network with human-provided attention guidance on where in the image the network should look for answering clues. On two orthopedics radiographic fracture classification datasets, through extensive experiments we demonstrate that explicit human-guided attention indeed can direct correct network attention and consequently significantly improve classification performance. The development code for the proposed attention guidance is publicly available on GitHub., Comment: 12 pages, Published in IEEE Journal of Biomedical and Health Informatics

The promise of active learning (AL) is to reduce labelling costs by selecting the most valuable examples to annotate from a pool of unlabelled data. Identifying these examples is especially challenging with high-dimensional data (e.g. images, videos) and in low-data regimes. In this paper, we propose a novel method for batch AL called ALFA-Mix. We identify unlabelled instances with sufficiently-distinct features by seeking inconsistencies in predictions resulting from interventions on their representations. We construct interpolations between representations of labelled and unlabelled instances then examine the predicted labels. We show that inconsistencies in these predictions help discovering features that the model is unable to recognise in the unlabelled instances. We derive an efficient implementation based on a closed-form solution to the optimal interpolation causing changes in predictions. Our method outperforms all recent AL approaches in 30 different settings on 12 benchmarks of images, videos, and non-visual data. The improvements are especially significant in low-data regimes and on self-trained vision transformers, where ALFA-Mix outperforms the state-of-the-art in 59% and 43% of the experiments respectively., Comment: CVPR 2022

The study of robustness has received much attention due to its inevitability in data-driven settings where many systems face uncertainty. One such example of concern is Bayesian Optimization (BO), where uncertainty is multi-faceted, yet there only exists a limited number of works dedicated to this direction. In particular, there is the work of Kirschner et al. (2020), which bridges the existing literature of Distributionally Robust Optimization (DRO) by casting the BO problem from the lens of DRO. While this work is pioneering, it admittedly suffers from various practical shortcomings such as finite contexts assumptions, leaving behind the main question Can one devise a computationally tractable algorithm for solving this DRO-BO problem? In this work, we tackle this question to a large degree of generality by considering robustness against data-shift in $\varphi$-divergences, which subsumes many popular choices, such as the $\chi^2$-divergence, Total Variation, and the extant Kullback-Leibler (KL) divergence. We show that the DRO-BO problem in this setting is equivalent to a finite-dimensional optimization problem which, even in the continuous context setting, can be easily implemented with provable sublinear regret bounds. We then show experimentally that our method surpasses existing methods, attesting to the theoretical results., Comment: NeurIPS 2023 camera ready paper

We introduce Retrieval Augmented Classification (RAC), a generic approach to augmenting standard image classification pipelines with an explicit retrieval module. RAC consists of a standard base image encoder fused with a parallel retrieval branch that queries a non-parametric external memory of pre-encoded images and associated text snippets. We apply RAC to the problem of long-tail classification and demonstrate a significant improvement over previous state-of-the-art on Places365-LT and iNaturalist-2018 (14.5% and 6.7% respectively), despite using only the training datasets themselves as the external information source. We demonstrate that RAC's retrieval module, without prompting, learns a high level of accuracy on tail classes. This, in turn, frees the base encoder to focus on common classes, and improve its performance thereon. RAC represents an alternative approach to utilizing large, pretrained models without requiring fine-tuning, as well as a first step towards more effectively making use of external memory within common computer vision architectures.

Continual Learning (CL) methods aim to enable machine learning models to learn new tasks without catastrophic forgetting of those that have been previously mastered. Existing CL approaches often keep a buffer of previously-seen samples, perform knowledge distillation, or use regularization techniques towards this goal. Despite their performance, they still suffer from interference across tasks which leads to catastrophic forgetting. To ameliorate this problem, we propose to only activate and select sparse neurons for learning current and past tasks at any stage. More parameters space and model capacity can thus be reserved for the future tasks. This minimizes the interference between parameters for different tasks. To do so, we propose a Sparse neural Network for Continual Learning (SNCL), which employs variational Bayesian sparsity priors on the activations of the neurons in all layers. Full Experience Replay (FER) provides effective supervision in learning the sparse activations of the neurons in different layers. A loss-aware reservoir-sampling strategy is developed to maintain the memory buffer. The proposed method is agnostic as to the network structures and the task boundaries. Experiments on different datasets show that our approach achieves state-of-the-art performance for mitigating forgetting.

Automated hate speech detection is an important tool in combating the spread of hate speech, particularly in social media. Numerous methods have been developed for the task, including a recent proliferation of deep-learning based approaches. A variety of datasets have also been developed, exemplifying various manifestations of the hate-speech detection problem. We present here a large-scale empirical comparison of deep and shallow hate-speech detection methods, mediated through the three most commonly used datasets. Our goal is to illuminate progress in the area, and identify strengths and weaknesses in the current state-of-the-art. We particularly focus our analysis on measures of practical performance, including detection accuracy, computational efficiency, capability in using pre-trained models, and domain generalization. In doing so we aim to provide guidance as to the use of hate-speech detection in practice, quantify the state-of-the-art, and identify future research directions. Code and dataset are available at https://github.com/jmjmalik22/Hate-Speech-Detection., Comment: 18 pages, 4 figures, and 6 tables

We propose an approach to semantic segmentation that achieves state-of-the-art supervised performance when applied in a zero-shot setting. It thus achieves results equivalent to those of the supervised methods, on each of the major semantic segmentation datasets, without training on those datasets. This is achieved by replacing each class label with a vector-valued embedding of a short paragraph that describes the class. The generality and simplicity of this approach enables merging multiple datasets from different domains, each with varying class labels and semantics. The resulting merged semantic segmentation dataset of over 2 Million images enables training a model that achieves performance equal to that of state-of-the-art supervised methods on 7 benchmark datasets, despite not using any images therefrom. By fine-tuning the model on standard semantic segmentation datasets, we also achieve a significant improvement over the state-of-the-art supervised segmentation on NYUD-V2 and PASCAL-context at 60% and 65% mIoU, respectively. Based on the closeness of language embeddings, our method can even segment unseen labels. Extensive experiments demonstrate strong generalization to unseen image domains and unseen labels, and that the method enables impressive performance improvements in downstream applications, including depth estimation and instance segmentation., Comment: 14 pages. Accepted to Int. J. Comp. Vis. (IJCV)

We propose a direct, regression-based approach to 2D human pose estimation from single images. We formulate the problem as a sequence prediction task, which we solve using a Transformer network. This network directly learns a regression mapping from images to the keypoint coordinates, without resorting to intermediate representations such as heatmaps. This approach avoids much of the complexity associated with heatmap-based approaches. To overcome the feature misalignment issues of previous regression-based methods, we propose an attention mechanism that adaptively attends to the features that are most relevant to the target keypoints, considerably improving the accuracy. Importantly, our framework is end-to-end differentiable, and naturally learns to exploit the dependencies between keypoints. Experiments on MS-COCO and MPII, two predominant pose-estimation datasets, demonstrate that our method significantly improves upon the state-of-the-art in regression-based pose estimation. More notably, ours is the first regression-based approach to perform favorably compared to the best heatmap-based pose estimation methods., Comment: Accepted to Proc. Eur. Conf. Comp. Vision (ECCV) 2022

Graph-level anomaly detection (GAD) describes the problem of detecting graphs that are abnormal in their structure and/or the features of their nodes, as compared to other graphs. One of the challenges in GAD is to devise graph representations that enable the detection of both locally- and globally-anomalous graphs, i.e., graphs that are abnormal in their fine-grained (node-level) or holistic (graph-level) properties, respectively. To tackle this challenge we introduce a novel deep anomaly detection approach for GAD that learns rich global and local normal pattern information by joint random distillation of graph and node representations. The random distillation is achieved by training one GNN to predict another GNN with randomly initialized network weights. Extensive experiments on 16 real-world graph datasets from diverse domains show that our model significantly outperforms seven state-of-the-art models. Code and datasets are available at https://git.io/GLocalKD., Comment: Accepted to WSDM 2022

Existing anomaly detection paradigms overwhelmingly focus on training detection models using exclusively normal data or unlabeled data (mostly normal samples). One notorious issue with these approaches is that they are weak in discriminating anomalies from normal samples due to the lack of the knowledge about the anomalies. Here, we study the problem of few-shot anomaly detection, in which we aim at using a few labeled anomaly examples to train sample-efficient discriminative detection models. To address this problem, we introduce a novel weakly-supervised anomaly detection framework to train detection models without assuming the examples illustrating all possible classes of anomaly. Specifically, the proposed approach learns discriminative normality (regularity) by leveraging the labeled anomalies and a prior probability to enforce expressive representations of normality and unbounded deviated representations of abnormality. This is achieved by an end-to-end optimization of anomaly scores with a neural deviation learning, in which the anomaly scores of normal samples are imposed to approximate scalar scores drawn from the prior while that of anomaly examples is enforced to have statistically significant deviations from these sampled scores in the upper tail. Furthermore, our model is optimized to learn fine-grained normality and abnormality by top-K multiple-instance-learning-based feature subspace deviation learning, allowing more generalized representations. Comprehensive experiments on nine real-world image anomaly detection benchmarks show that our model is substantially more sample-efficient and robust, and performs significantly better than state-of-the-art competing methods in both closed-set and open-set settings. Our model can also offer explanation capability as a result of its prior-driven anomaly score learning. Code and datasets are available at: https://git.io/DevNet., Comment: 16 pages, 8 figures, 5 tables

We propose an approach to instance segmentation from 3D point clouds based on dynamic convolution. This enables it to adapt, at inference, to varying feature and object scales. Doing so avoids some pitfalls of bottom up approaches, including a dependence on hyper-parameter tuning and heuristic post-processing pipelines to compensate for the inevitable variability in object sizes, even within a single scene. The representation capability of the network is greatly improved by gathering homogeneous points that have identical semantic categories and close votes for the geometric centroids. Instances are then decoded via several simple convolution layers, where the parameters are generated conditioned on the input. The proposed approach is proposal-free, and instead exploits a convolution process that adapts to the spatial and semantic characteristics of each instance. A light-weight transformer, built on the bottleneck layer, allows the model to capture long-range dependencies, with limited computational overhead. The result is a simple, efficient, and robust approach that yields strong performance on various datasets: ScanNetV2, S3DIS, and PartNet. The consistent improvements on both voxel- and point-based architectures imply the effectiveness of the proposed method. Code is available at: https://git.io/DyCo3D, Comment: Accepted to IEEE Trans. Pattern Analysis and Machine Intelligence. Extended version of arXiv:2011.13328

Neural networks trained with SGD were recently shown to rely preferentially on linearly-predictive features and can ignore complex, equally-predictive ones. This simplicity bias can explain their lack of robustness out of distribution (OOD). The more complex the task to learn, the more likely it is that statistical artifacts (i.e. selection biases, spurious correlations) are simpler than the mechanisms to learn. We demonstrate that the simplicity bias can be mitigated and OOD generalization improved. We train a set of similar models to fit the data in different ways using a penalty on the alignment of their input gradients. We show theoretically and empirically that this induces the learning of more complex predictive patterns. OOD generalization fundamentally requires information beyond i.i.d. examples, such as multiple training environments, counterfactual examples, or other side information. Our approach shows that we can defer this requirement to an independent model selection stage. We obtain SOTA results in visual recognition on biased data and generalization across visual domains. The method - the first to evade the simplicity bias - highlights the need for a better understanding and control of inductive biases in deep learning., Comment: CVPR 2022

Vision-and-Language Navigation (VLN) requires an agent to find a path to a remote location on the basis of natural-language instructions and a set of photo-realistic panoramas. Most existing methods take the words in the instructions and the discrete views of each panorama as the minimal unit of encoding. However, this requires a model to match different nouns (e.g., TV, table) against the same input view feature. In this work, we propose an object-informed sequential BERT to encode visual perceptions and linguistic instructions at the same fine-grained level, namely objects and words. Our sequential BERT also enables the visual-textual clues to be interpreted in light of the temporal context, which is crucial to multi-round VLN tasks. Additionally, we enable the model to identify the relative direction (e.g., left/right/front/back) of each navigable location and the room type (e.g., bedroom, kitchen) of its current and final navigation goal, as such information is widely mentioned in instructions implying the desired next and final locations. We thus enable the model to know-where the objects lie in the images, and to know-where they stand in the scene. Extensive experiments demonstrate the effectiveness compared against several state-of-the-art methods on three indoor VLN tasks: REVERIE, NDH, and R2R. Project repository: https://github.com/YuankaiQi/ORIST, Comment: Original title: Know What and Know Where: An Object-and-Room Informed Sequential BERT for Indoor Vision-Language Navigation

Visual navigation is often cast as a reinforcement learning (RL) problem. Current methods typically result in a suboptimal policy that learns general obstacle avoidance and search behaviours. For example, in the target-object navigation setting, the policies learnt by traditional methods often fail to complete the task, even when the target is clearly within reach from a human perspective. In order to address this issue, we propose to learn to imagine a latent representation of the successful (sub-)goal state. To do so, we have developed a module which we call Foresight Imagination (ForeSIT). ForeSIT is trained to imagine the recurrent latent representation of a future state that leads to success, e.g. either a sub-goal state that is important to reach before the target, or the goal state itself. By conditioning the policy on the generated imagination during training, our agent learns how to use this imagination to achieve its goal robustly. Our agent is able to imagine what the (sub-)goal state may look like (in the latent space) and can learn to navigate towards that state. We develop an efficient learning algorithm to train ForeSIT in an on-policy manner and integrate it into our RL objective. The integration is not trivial due to the constantly evolving state representation shared between both the imagination and the policy. We, empirically, observe that our method outperforms the state-of-the-art methods by a large margin in the commonly accepted benchmark AI2THOR environment. Our method can be readily integrated or added to other model-free RL navigation frameworks.

The limits of applicability of vision-and-language models are defined by the coverage of their training data. Tasks like vision question answering (VQA) often require commonsense and factual information beyond what can be learned from task-specific datasets. This paper investigates the injection of knowledge from general-purpose knowledge bases (KBs) into vision-and-language transformers. We use an auxiliary training objective that encourages the learned representations to align with graph embeddings of matching entities in a KB. We empirically study the relevance of various KBs to multiple tasks and benchmarks. The technique brings clear benefits to knowledge-demanding question answering tasks (OK-VQA, FVQA) by capturing semantic and relational knowledge absent from existing models. More surprisingly, the technique also benefits visual reasoning tasks (NLVR2, SNLI-VE). We perform probing experiments and show that the injection of additional knowledge regularizes the space of embeddings, which improves the representation of lexical and semantic similarities. The technique is model-agnostic and can expand the applicability of any vision-and-language transformer with minimal computational overhead.

A wide variety of methods have been developed for identifying depression, but they focus primarily on measuring the degree to which individuals are suffering from depression currently. In this work we explore the possibility of predicting future depression using machine learning applied to longitudinal socio-demographic data. In doing so we show that data such as housing status, and the details of the family environment, can provide cues for predicting future psychiatric disorders. To this end, we introduce a novel deep multi-task recurrent neural network to learn time-dependent depression cues. The depression prediction task is jointly optimized with two auxiliary anomaly ranking tasks, including contrastive one-class feature ranking and deviation ranking. The auxiliary tasks address two key challenges of the problem: 1) the high within class variance of depression samples: they enable the learning of representations that are robust to highly variant in-class distribution of the depression samples; and 2) the small labeled data volume: they significantly enhance the sample efficiency of the prediction model, which reduces the reliance on large depression-labeled datasets that are difficult to collect in practice. Extensive empirical results on large-scale child depression data show that our model is sample-efficient and can accurately predict depression 2-4 years before the illness occurs, substantially outperforming eight representative comparators., Comment: Accepted to PAKDD 2022

Previous top-performing approaches for point cloud instance segmentation involve a bottom-up strategy, which often includes inefficient operations or complex pipelines, such as grouping over-segmented components, introducing additional steps for refining, or designing complicated loss functions. The inevitable variation in the instance scales can lead bottom-up methods to become particularly sensitive to hyper-parameter values. To this end, we propose instead a dynamic, proposal-free, data-driven approach that generates the appropriate convolution kernels to apply in response to the nature of the instances. To make the kernels discriminative, we explore a large context by gathering homogeneous points that share identical semantic categories and have close votes for the geometric centroids. Instances are then decoded by several simple convolutional layers. Due to the limited receptive field introduced by the sparse convolution, a small light-weight transformer is also devised to capture the long-range dependencies and high-level interactions among point samples. The proposed method achieves promising results on both ScanetNetV2 and S3DIS, and this performance is robust to the particular hyper-parameter values chosen. It also improves inference speed by more than 25% over the current state-of-the-art. Code is available at: https://git.io/DyCo3D, Comment: Appearing in IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2021

We consider the problem of anomaly detection with a small set of partially labeled anomaly examples and a large-scale unlabeled dataset. This is a common scenario in many important applications. Existing related methods either exclusively fit the limited anomaly examples that typically do not span the entire set of anomalies, or proceed with unsupervised learning from the unlabeled data. We propose here instead a deep reinforcement learning-based approach that enables an end-to-end optimization of the detection of both labeled and unlabeled anomalies. This approach learns the known abnormality by automatically interacting with an anomaly-biased simulation environment, while continuously extending the learned abnormality to novel classes of anomaly (i.e., unknown anomalies) by actively exploring possible anomalies in the unlabeled data. This is achieved by jointly optimizing the exploitation of the small labeled anomaly data and the exploration of the rare unlabeled anomalies. Extensive experiments on 48 real-world datasets show that our model significantly outperforms five state-of-the-art competing methods., Comment: Accepted to KDD 2021

Vision-and-Language Navigation (VLN) is unique in that it requires turning relatively general natural-language instructions into robot agent actions, on the basis of the visible environment. This requires to extract value from two very different types of natural-language information. The first is object description (e.g., 'table', 'door'), each presenting as a tip for the agent to determine the next action by finding the item visible in the environment, and the second is action specification (e.g., 'go straight', 'turn left') which allows the robot to directly predict the next movements without relying on visual perceptions. However, most existing methods pay few attention to distinguish these information from each other during instruction encoding and mix together the matching between textual object/action encoding and visual perception/orientation features of candidate viewpoints. In this paper, we propose an Object-and-Action Aware Model (OAAM) that processes these two different forms of natural language based instruction separately. This enables each process to match object-centered/action-centered instruction to their own counterpart visual perception/action orientation flexibly. However, one side-issue caused by above solution is that an object mentioned in instructions may be observed in the direction of two or more candidate viewpoints, thus the OAAM may not predict the viewpoint on the shortest path as the next action. To handle this problem, we design a simple but effective path loss to penalize trajectories deviating from the ground truth path. Experimental results demonstrate the effectiveness of the proposed model and path loss, and the superiority of their combination with a 50% SPL score on the R2R dataset and a 40% CLS score on the R4R dataset in unseen environments, outperforming the previous state-of-the-art.

Anomaly detection, a.k.a. outlier detection or novelty detection, has been a lasting yet active research area in various research communities for several decades. There are still some unique problem complexities and challenges that require advanced approaches. In recent years, deep learning enabled anomaly detection, i.e., deep anomaly detection, has emerged as a critical direction. This paper surveys the research of deep anomaly detection with a comprehensive taxonomy, covering advancements in three high-level categories and 11 fine-grained categories of the methods. We review their key intuitions, objective functions, underlying assumptions, advantages and disadvantages, and discuss how they address the aforementioned challenges. We further discuss a set of possible future opportunities and new perspectives on addressing the challenges., Comment: Survey paper, 36 pages, 180 references, 2 figures, 4 tables