Christian Gagné

Arman Afrasiyabi, Hugo Larochelle, Jean-François Lalonde et al.

In image classification, it is common practice to train deep networks to extract a single feature vector per input image. Few-shot classification methods also mostly follow this trend. In this work, we depart from this established direction and instead propose to extract sets of feature vectors for each image. We argue that a set-based representation intrinsically builds a richer representation of images from the base classes, which can subsequently better transfer to the few-shot classes. To do so, we propose to adapt existing feature extractors to instead produce sets of feature vectors from images. Our approach, dubbed SetFeat, embeds shallow self-attention mechanisms inside existing encoder architectures. The attention modules are lightweight, and as such our method results in encoders that have approximately the same number of parameters as their original versions. During training and inference, a set-to-set matching metric is used to perform image classification. The effectiveness of our proposed architecture and metrics is demonstrated via thorough experiments on standard few-shot datasets -- namely miniImageNet, tieredImageNet, and CUB -- in both the 1- and 5-shot scenarios. In all cases but one, our method outperforms the state-of-the-art.

Jiaqi Li, Yuanhao Lai, Rui Wang et al.

Continual learning aims to learn a series of tasks sequentially without forgetting the knowledge acquired from the previous ones. In this work, we propose the Hessian Aware Low-Rank Perturbation algorithm for continual learning. By modeling the parameter transitions along the sequential tasks with the weight matrix transformation, we propose to apply the low-rank approximation on the task-adaptive parameters in each layer of the neural networks. Specifically, we theoretically demonstrate the quantitative relationship between the Hessian and the proposed low-rank approximation. The approximation ranks are then globally determined according to the marginal increment of the empirical loss estimated by the layer-specific gradient and low-rank approximation error. Furthermore, we control the model capacity by pruning less important parameters to diminish the parameter growth. We conduct extensive experiments on various benchmarks, including a dataset with large-scale tasks, and compare our method against some recent state-of-the-art methods to demonstrate the effectiveness and scalability of our proposed method. Empirical results show that our method performs better on different benchmarks, especially in achieving task order robustness and handling the forgetting issue. The source code is at https://github.com/lijiaqi/HALRP.

Changjian Shui, Gezheng Xu, Qi Chen et al.

We propose an analysis in fair learning that preserves the utility of the data while reducing prediction disparities under the criteria of group sufficiency. We focus on the scenario where the data contains multiple or even many subgroups, each with limited number of samples. As a result, we present a principled method for learning a fair predictor for all subgroups via formulating it as a bilevel objective. Specifically, the subgroup specific predictors are learned in the lower-level through a small amount of data and the fair predictor. In the upper-level, the fair predictor is updated to be close to all subgroup specific predictors. We further prove that such a bilevel objective can effectively control the group sufficiency and generalization error. We evaluate the proposed framework on real-world datasets. Empirical evidence suggests the consistently improved fair predictions, as well as the comparable accuracy to the baselines.

Thomas Philippon, Christian Gagné

Neural network ensembles have been studied extensively in the context of adversarial robustness and most ensemble-based approaches remain vulnerable to adaptive attacks. In this paper, we investigate the robustness of Error-Correcting Output Codes (ECOC) ensembles through architectural improvements and ensemble diversity promotion. We perform a comprehensive robustness assessment against adaptive attacks and investigate the relationship between ensemble diversity and robustness. Our results demonstrate the benefits of ECOC ensembles for adversarial robustness compared to regular ensembles of convolutional neural networks (CNNs) and show why the robustness of previous implementations is limited. We also propose an adversarial training method specific to ECOC ensembles that allows to further improve robustness to adaptive attacks.

Changjian Shui, Qi Chen, Jiaqi Li et al.

We consider a fair representation learning perspective, where optimal predictors, on top of the data representation, are ensured to be invariant with respect to different sub-groups. Specifically, we formulate this intuition as a bi-level optimization, where the representation is learned in the outer-loop, and invariant optimal group predictors are updated in the inner-loop. Moreover, the proposed bi-level objective is demonstrated to fulfill the sufficiency rule, which is desirable in various practical scenarios but was not commonly studied in the fair learning. Besides, to avoid the high computational and memory cost of differentiating in the inner-loop of bi-level objective, we propose an implicit path alignment algorithm, which only relies on the solution of inner optimization and the implicit differentiation rather than the exact optimization path. We further analyze the error gap of the implicit approach and empirically validate the proposed method in both classification and regression settings. Experimental results show the consistently better trade-off in prediction performance and fairness measurement.

William Wei Wang, Gezheng Xu, Ruizhi Pu et al.

Domain generalization aims to learn a predictive model from multiple different but related source tasks that can generalize well to a target task without the need of accessing any target data. Existing domain generalization methods ignore the relationship between tasks, implicitly assuming that all the tasks are sampled from a stationary environment. Therefore, they can fail when deployed in an evolving environment. To this end, we formulate and study the \emph{evolving domain generalization} (EDG) scenario, which exploits not only the source data but also their evolving pattern to generate a model for the unseen task. Our theoretical result reveals the benefits of modeling the relation between two consecutive tasks by learning a globally consistent directional mapping function. In practice, our analysis also suggests solving the DDG problem in a meta-learning manner, which leads to \emph{directional prototypical network}, the first method for the DDG problem. Empirical evaluation of both synthetic and real-world data sets validates the effectiveness of our approach.

Fatemeh Nourilenjan Nokabadi, Yann Batiste Pequignot, Jean-Francois Lalonde et al.

Adversarial perturbations can deceive neural networks by adding small, imperceptible noise to the input. Recent object trackers with transformer backbones have shown strong performance on tracking datasets, but their adversarial robustness has not been thoroughly evaluated. While transformer trackers are resilient to black-box attacks, existing white-box adversarial attacks are not universally applicable against these new transformer trackers due to differences in backbone architecture. In this work, we introduce TrackPGD, a novel white-box attack that utilizes predicted object binary masks to target robust transformer trackers. Built upon the powerful segmentation attack SegPGD, our proposed TrackPGD effectively influences the decisions of transformer-based trackers. Our method addresses two primary challenges in adapting a segmentation attack for trackers: limited class numbers and extreme pixel class imbalance. TrackPGD uses the same number of iterations as other attack methods for tracker networks and produces competitive adversarial examples that mislead transformer and non-transformer trackers such as MixFormerM, OSTrackSTS, TransT-SEG, and RTS on datasets including VOT2022STS, DAVIS2016, UAV123, and GOT-10k.

Vikash Singh, Debargha Ganguly, Weicong Chen et al.

Continual test-time adaptation (CTTA) updates a pretrained model online on an unlabeled, non-stationary stream while anchoring it to a frozen source checkpoint. This anchor is useful only when the source remains reliable. On CCC-Hard, however, a ResNet-50 source falls to approximately $1.3\%$ top-$1$ accuracy, while existing source-anchored CTTA methods continue applying the same anchor strength. We call this failure mode blind anchoring and propose RMemSafe, a reliability-gated extension of ROID that uses the frozen source's normalized predictive entropy to attenuate all explicit source-coupled uses in the objective. When the source posterior approaches uniformity, the gate closes: the source anchor and agreement filter vanish, and the objective reduces to a source-agnostic fallback comprising ROID's base losses plus marginal calibration. Combined with ASR, RMemSafe achieves the lowest error on $8$ of $9$ matched-split continual-corruption cells and is the best reset-based method on all $9$, improving ROID+ASR by $1.05$~pp on ResNet-50 and $0.48$~pp on ViT-B/16. A controlled source-degradation sweep shows a $1.13{\times}$ shallower harm slope than ROID+ASR, consistent with the graceful-decay prediction. The entropy gate detects high-entropy source collapse, not confidently wrong low-entropy sources; this scope is explicitly evaluated and discussed.

Jonas Ngnawé, Maxime Heuillet, Sabyasachi Sahoo et al.

Fine-tuning pretrained models is a standard and effective workflow in modern machine learning. However, robust fine-tuning (RFT), which aims to simultaneously achieve adaptation to a downstream task and robustness to adversarial examples, remains challenging. Despite the abundance of non-robust pretrained models in open-source repositories, their potential for RFT is less understood. We address this knowledge gap by systematically examining RFT from such non-robust models. Our experiments reveal that fine-tuning non-robust models with a robust objective, even under small perturbations, can lead to poor performance, a phenomenon that we dub \emph{suboptimal transfer}. In challenging scenarios (eg, difficult tasks, high perturbation), the resulting performance can be so low that it may be considered a transfer failure. We find that fine-tuning using a robust objective impedes task adaptation at the beginning of training and eventually prevents optimal transfer. However, we propose a novel heuristic, \emph{Epsilon-Scheduling}, a schedule over perturbation strength used during training that promotes optimal transfer. Additionally, we introduce \emph{expected robustness}, a metric that captures performance across a range of perturbations, providing a more comprehensive evaluation of the accuracy-robustness trade-off for diverse models at test time. Extensive experiments on a wide range of configurations (six pretrained models and five datasets) show that \emph{Epsilon-Scheduling} successfully prevents \emph{suboptimal transfer} and consistently improves expected robustness.

Fatemeh Nourilenjan Nokabadi, Jean-François Lalonde, Christian Gagné

New transformer networks have been integrated into object tracking pipelines and have demonstrated strong performance on the latest benchmarks. This paper focuses on understanding how transformer trackers behave under adversarial attacks and how different attacks perform on tracking datasets as their parameters change. We conducted a series of experiments to evaluate the effectiveness of existing adversarial attacks on object trackers with transformer and non-transformer backbones. We experimented on 7 different trackers, including 3 that are transformer-based, and 4 which leverage other architectures. These trackers are tested against 4 recent attack methods to assess their performance and robustness on VOT2022ST, UAV123 and GOT10k datasets. Our empirical study focuses on evaluating adversarial robustness of object trackers based on bounding box versus binary mask predictions, and attack methods at different levels of perturbations. Interestingly, our study found that altering the perturbation level may not significantly affect the overall object tracking results after the attack. Similarly, the sparsity and imperceptibility of the attack perturbations may remain stable against perturbation level shifts. By applying a specific attack on all transformer trackers, we show that new transformer trackers having a stronger cross-attention modeling achieve a greater adversarial robustness on tracking datasets, such as VOT2022ST and GOT10k. Our results also indicate the necessity for new attack methods to effectively tackle the latest types of transformer trackers. The codes necessary to reproduce this study are available at https://github.com/fatemehN/ReproducibilityStudy.

Maxime Heuillet, Rishika Bhagwatkar, Jonas Ngnawé et al.

Deep learning models operating in the image domain are vulnerable to small input perturbations. For years, robustness to such perturbations was pursued by training models from scratch (i.e., with random initializations) using specialized loss objectives. Recently, robust fine-tuning has emerged as a more efficient alternative: instead of training from scratch, pretrained models are adapted to maximize predictive performance and robustness. To conduct robust fine-tuning, practitioners design an optimization strategy that includes the model update protocol (e.g., full or partial) and the specialized loss objective. Additional design choices include the architecture type and size, and the pretrained representation. These design choices affect robust generalization, which is the model's ability to maintain performance when exposed to new and unseen perturbations at test time. Understanding how these design choices influence generalization remains an open question with significant practical implications. In response, we present an empirical study spanning 6 datasets, 40 pretrained architectures, 2 specialized losses, and 3 adaptation protocols, yielding 1,440 training configurations and 7,200 robustness measurements across five perturbation types. To our knowledge, this is the most diverse and comprehensive benchmark of robust fine-tuning to date. While attention-based architectures and robust pretrained representations are increasingly popular, we find that convolutional neural networks pretrained in a supervised manner on large datasets often perform best. Our analysis both confirms and challenges prior design assumptions, highlighting promising research directions and offering practical guidance.

Fatemeh Nourilenjan Nokabadi, Jean-Francois Lalonde, Christian Gagné

Adversarial perturbations aim to deceive neural networks into predicting inaccurate results. For visual object trackers, adversarial attacks have been developed to generate perturbations by manipulating the outputs. However, transformer trackers predict a specific bounding box instead of an object candidate list, which limits the applicability of many existing attack scenarios. To address this issue, we present a novel white-box approach to attack visual object trackers with transformer backbones using only one bounding box. From the tracker predicted bounding box, we generate a list of adversarial bounding boxes and compute the adversarial loss for those bounding boxes. Experimental results demonstrate that our simple yet effective attack outperforms existing attacks against several robust transformer trackers, including TransT-M, ROMTrack, and MixFormer, on popular benchmark tracking datasets such as GOT-10k, UAV123, and VOT2022STS.

Adam Tupper, Christian Gagné

Foundation models open up new possibilities for the use of AI in healthcare. However, even when pre-trained on health data, they still need to be fine-tuned for specific downstream tasks. Furthermore, although foundation models reduce the amount of training data required to achieve good performance, obtaining sufficient data is still a challenge. This is due, in part, to restrictions on sharing and aggregating data from different sources to protect patients' privacy. One possible solution to this is to fine-tune foundation models via federated learning across multiple participating clients (i.e., hospitals, clinics, etc.). In this work, we propose a new personalized federated fine-tuning method that learns orthogonal LoRA adapters to disentangle general and client-specific knowledge, enabling each client to fully exploit both their own data and the data of others. Our preliminary results on real-world federated medical imaging tasks demonstrate that our approach is competitive against current federated fine-tuning methods.

Róisín Luo, James McDermott, Christian Gagné et al.

Lipschitz continuity characterizes the worst-case sensitivity of neural networks to small input perturbations; yet its dynamics (i.e. temporal evolution) during training remains under-explored. We present a rigorous mathematical framework to model the temporal evolution of Lipschitz continuity during training with stochastic gradient descent (SGD). This framework leverages a system of stochastic differential equations (SDEs) to capture both deterministic and stochastic forces. Our theoretical analysis identifies three principal factors driving the evolution: (i) the projection of gradient flows, induced by the optimization dynamics, onto the operator-norm Jacobian of parameter matrices; (ii) the projection of gradient noise, arising from the randomness in mini-batch sampling, onto the operator-norm Jacobian; and (iii) the projection of the gradient noise onto the operator-norm Hessian of parameter matrices. Furthermore, our theoretical framework sheds light on such as how noisy supervision, parameter initialization, batch size, and mini-batch sampling trajectories, among other factors, shape the evolution of the Lipschitz continuity of neural networks. Our experimental results demonstrate strong agreement between the theoretical implications and the observed behaviors.

Adam Tupper, Christian Gagné

Data augmentation is a widely used and effective technique to improve the generalization performance of deep neural networks. Yet, despite often facing limited data availability when working with medical images, it is frequently underutilized. This appears to come from a gap in our collective understanding of the efficacy of different augmentation techniques across different tasks and modalities. One modality where this is especially true is ultrasound imaging. This work addresses this gap by analyzing the effectiveness of different augmentation techniques at improving model performance across a wide range of ultrasound image analysis tasks. To achieve this, we introduce a new standardized benchmark of 14 ultrasound image classification and semantic segmentation tasks from 10 different sources and covering 11 body regions. Our results demonstrate that many of the augmentations commonly used for tasks on natural images are also effective on ultrasound images, even more so than augmentations developed specifically for ultrasound images in some cases. We also show that diverse augmentation using TrivialAugment, which is widely used for natural images, is also effective for ultrasound images. Moreover, our proposed methodology represents a structured approach for assessing various data augmentations that can be applied to other contexts and modalities.

Jonas Ngnawé, Sabyasachi Sahoo, Yann Pequignot et al.

Despite extensive research on adversarial training strategies to improve robustness, the decisions of even the most robust deep learning models can still be quite sensitive to imperceptible perturbations, creating serious risks when deploying them for high-stakes real-world applications. While detecting such cases may be critical, evaluating a model's vulnerability at a per-instance level using adversarial attacks is computationally too intensive and unsuitable for real-time deployment scenarios. The input space margin is the exact score to detect non-robust samples and is intractable for deep neural networks. This paper introduces the concept of margin consistency -- a property that links the input space margins and the logit margins in robust models -- for efficient detection of vulnerable samples. First, we establish that margin consistency is a necessary and sufficient condition to use a model's logit margin as a score for identifying non-robust samples. Next, through comprehensive empirical analysis of various robustly trained models on CIFAR10 and CIFAR100 datasets, we show that they indicate high margin consistency with a strong correlation between their input space margins and the logit margins. Then, we show that we can effectively and confidently use the logit margin to detect brittle decisions with such models. Finally, we address cases where the model is not sufficiently margin-consistent by learning a pseudo-margin from the feature representation. Our findings highlight the potential of leveraging deep representations to assess adversarial vulnerability in deployment scenarios efficiently.

Adam Tupper, Christian Gagné

Data augmentation is one of the most effective techniques to improve the generalization performance of deep neural networks. Yet, despite often facing limited data availability in medical image analysis, it is frequently underutilized. This appears to be due to a gap in our collective understanding of the efficacy of different augmentation techniques across medical imaging tasks and modalities. One domain where this is especially true is breast ultrasound images. This work addresses this issue by analyzing the effectiveness of different augmentation techniques for the classification of breast lesions in ultrasound images. We assess the generalizability of our findings across several datasets, demonstrate that certain augmentations are far more effective than others, and show that their usage leads to significant performance gains.

Catherine Bouchard, Andréanne Deschênes, Vincent Boulanger et al.

The development of signal unmixing algorithms is essential for leveraging multimodal datasets acquired through a wide array of scientific imaging technologies, including hyperspectral or time-resolved acquisitions. In experimental physics, enhancing the spatio-temporal resolution or expanding the number of detection channels often leads to diminished sampling rate and signal-to-noise ratio, significantly affecting the efficacy of signal unmixing algorithms. We propose Latent Unmixing, a new approach which applies bandpass filters to the latent space of a multidimensional convolutional neural network to disentangle overlapping signal components. It enables better isolation and quantification of individual signal contributions, especially in the context of undersampled distributions. Using multidimensional convolution kernels to process all dimensions simultaneously enhances the network's ability to extract information from adjacent pixels, and time or spectral bins. This approach enables more effective separation of components in cases where individual pixels do not provide clear, well-resolved information. We showcase the method's practical use in experimental physics through two test cases that highlight the versatility of our approach: fluorescence lifetime microscopy and mode decomposition in optical fibers. The latent unmixing method extracts valuable information from complex signals that cannot be resolved by standard methods. It opens up new possibilities in optics and photonics for multichannel separation at an increased sampling rate.

Mohamed Abid, Arman Afrasiyabi, Ihsen Hedhli et al.

Generally, image-to-image translation (i2i) methods aim at learning mappings across domains with the assumption that the images used for translation share content (e.g., pose) but have their own domain-specific information (a.k.a. style). Conditioned on a target image, such methods extract the target style and combine it with the source image content, keeping coherence between the domains. In our proposal, we depart from this traditional view and instead consider the scenario where the target domain is represented by a very low-resolution (LR) image, proposing a domain-agnostic i2i method for fine-grained problems, where the domains are related. More specifically, our domain-agnostic approach aims at generating an image that combines visual features from the source image with low-frequency information (e.g. pose, color) of the LR target image. To do so, we present a novel approach that relies on training the generative model to produce images that both share distinctive information of the associated source image and correctly match the LR target image when downscaled. We validate our method on the CelebA-HQ and AFHQ datasets by demonstrating improvements in terms of visual quality. Qualitative and quantitative results show that when dealing with intra-domain image translation, our method generates realistic samples compared to state-of-the-art methods such as StarGAN v2. Ablation studies also reveal that our method is robust to changes in color, it can be applied to out-of-distribution images, and it allows for manual control over the final results.

Boyu Wang, Jorge Mendez, Changjian Shui et al.

Learning from multiple related tasks by knowledge sharing and transfer has become increasingly relevant over the last two decades. In order to successfully transfer information from one task to another, it is critical to understand the similarities and differences between the domains. In this paper, we introduce the notion of \emph{performance gap}, an intuitive and novel measure of the distance between learning tasks. Unlike existing measures which are used as tools to bound the difference of expected risks between tasks (e.g., $\mathcal{H}$-divergence or discrepancy distance), we theoretically show that the performance gap can be viewed as a data- and algorithm-dependent regularizer, which controls the model complexity and leads to finer guarantees. More importantly, it also provides new insights and motivates a novel principle for designing strategies for knowledge sharing and transfer: gap minimization. We instantiate this principle with two algorithms: 1. gapBoost, a novel and principled boosting algorithm that explicitly minimizes the performance gap between source and target domains for transfer learning; and 2. gapMTNN, a representation learning algorithm that reformulates gap minimization as semantic conditional matching for multitask learning. Our extensive evaluation on both transfer learning and multitask learning benchmark data sets shows that our methods outperform existing baselines.

Changjian Shui, Boyu Wang, Christian Gagné

A crucial aspect in reliable machine learning is to design a deployable system in generalizing new related but unobserved environments. Domain generalization aims to alleviate such a prediction gap between the observed and unseen environments. Previous approaches commonly incorporated learning invariant representation for achieving good empirical performance. In this paper, we reveal that merely learning invariant representation is vulnerable to the unseen environment. To this end, we derive novel theoretical analysis to control the unseen test environment error in the representation learning, which highlights the importance of controlling the smoothness of representation. In practice, our analysis further inspires an efficient regularization method to improve the robustness in domain generalization. Our regularization is orthogonal to and can be straightforwardly adopted in existing domain generalization algorithms for invariant representation learning. Empirical results show that our algorithm outperforms the base versions in various dataset and invariance criteria.

Changjian Shui, Zijian Li, Jiaqi Li et al.

Multi-source domain adaptation aims at leveraging the knowledge from multiple tasks for predicting a related target domain. Hence, a crucial aspect is to properly combine different sources based on their relations. In this paper, we analyzed the problem for aggregating source domains with different label distributions, where most recent source selection approaches fail. Our proposed algorithm differs from previous approaches in two key ways: the model aggregates multiple sources mainly through the similarity of semantic conditional distribution rather than marginal distribution; the model proposes a \emph{unified} framework to select relevant sources for three popular scenarios, i.e., domain adaptation with limited label on target domain, unsupervised domain adaptation and label partial unsupervised domain adaption. We evaluate the proposed method through extensive experiments. The empirical results significantly outperform the baselines.

Hugo Siqueira Gomes, Benjamin Léger, Christian Gagné

The no free lunch theorem states that no model is better suited to every problem. A question that arises from this is how to design methods that propose optimizers tailored to specific problems achieving state-of-the-art performance. This paper addresses this issue by proposing the use of meta-learning to infer population-based black-box optimizers that can automatically adapt to specific classes of problems. We suggest a general modeling of population-based algorithms that result in Learning-to-Optimize POMDP (LTO-POMDP), a meta-learning framework based on a specific partially observable Markov decision process (POMDP). From that framework's formulation, we propose to parameterize the algorithm using deep recurrent neural networks and use a meta-loss function based on stochastic algorithms' performance to train efficient data-driven optimizers over several related optimization tasks. The learned optimizers' performance based on this implementation is assessed on various black-box optimization tasks and hyperparameter tuning of machine learning models. Our results revealed that the meta-loss function encourages a learned algorithm to alter its search behavior so that it can easily fit into a new context. Thus, it allows better generalization and higher sample efficiency than state-of-the-art generic optimization algorithms, such as the Covariance matrix adaptation evolution strategy (CMA-ES).

Mohamed Abderrahmen Abid, Ihsen Hedhli, Christian Gagné

Traditionally, the main focus of image super-resolution techniques is on recovering the most likely high-quality images from low-quality images, using a one-to-one low- to high-resolution mapping. Proceeding that way, we ignore the fact that there are generally many valid versions of high-resolution images that map to a given low-resolution image. We are tackling in this work the problem of obtaining different high-resolution versions from the same low-resolution image using Generative Adversarial Models. Our learning approach makes use of high frequencies available in the training high-resolution images for preserving and exploring in an unsupervised manner the structural information available within these images. Experimental results on the CelebA dataset confirm the effectiveness of the proposed method, which allows the generation of both realistic and diverse high-resolution images from low-resolution images.

Arman Afrasiyabi, Jean-François Lalonde, Christian Gagné

We introduce Mixture-based Feature Space Learning (MixtFSL) for obtaining a rich and robust feature representation in the context of few-shot image classification. Previous works have proposed to model each base class either with a single point or with a mixture model by relying on offline clustering algorithms. In contrast, we propose to model base classes with mixture models by simultaneously training the feature extractor and learning the mixture model parameters in an online manner. This results in a richer and more discriminative feature space which can be employed to classify novel examples from very few samples. Two main stages are proposed to train the MixtFSL model. First, the multimodal mixtures for each base class and the feature extractor parameters are learned using a combination of two loss functions. Second, the resulting network and mixture models are progressively refined through a leader-follower learning procedure, which uses the current estimate as a "target" network. This target network is used to make a consistent assignment of instances to mixture components, which increases performance and stabilizes training. The effectiveness of our end-to-end feature space learning approach is demonstrated with extensive experiments on four standard datasets and four backbones. Notably, we demonstrate that when we combine our robust representation with recent alignment-based approaches, we achieve new state-of-the-art results in the inductive setting, with an absolute accuracy for 5-shot classification of 82.45 on miniImageNet, 88.20 with tieredImageNet, and 60.70 in FC100 using the ResNet-12 backbone.

Changjian Shui, Qi Chen, Jun Wen et al.

We reveal the incoherence between the widely-adopted empirical domain adversarial training and its generally-assumed theoretical counterpart based on $\mathcal{H}$-divergence. Concretely, we find that $\mathcal{H}$-divergence is not equivalent to Jensen-Shannon divergence, the optimization objective in domain adversarial training. To this end, we establish a new theoretical framework by directly proving the upper and lower target risk bounds based on joint distributional Jensen-Shannon divergence. We further derive bi-directional upper bounds for marginal and conditional shifts. Our framework exhibits inherent flexibilities for different transfer learning problems, which is usable for various scenarios where $\mathcal{H}$-divergence-based theory fails to adapt. From an algorithmic perspective, our theory enables a generic guideline unifying principles of semantic conditional matching, feature marginal matching, and label marginal shift correction. We employ algorithms for each principle and empirically validate the benefits of our framework on real datasets.

Sébastien de Blois, Mathieu Garon, Christian Gagné et al.

Computer vision datasets containing multiple modalities such as color, depth, and thermal properties are now commonly accessible and useful for solving a wide array of challenging tasks. However, deploying multi-sensor heads is not possible in many scenarios. As such many practical solutions tend to be based on simpler sensors, mostly for cost, simplicity and robustness considerations. In this work, we propose a training methodology to take advantage of these additional modalities available in datasets, even if they are not available at test time. By assuming that the modalities have a strong spatial correlation, we propose Input Dropout, a simple technique that consists in stochastic hiding of one or many input modalities at training time, while using only the canonical (e.g. RGB) modalities at test time. We demonstrate that Input Dropout trivially combines with existing deep convolutional architectures, and improves their performance on a wide range of computer vision tasks such as dehazing, 6-DOF object tracking, pedestrian detection and object classification.

Arman Afrasiyabi, Jean-François Lalonde, Christian Gagné

Few-shot image classification aims at training a model from only a few examples for each of the "novel" classes. This paper proposes the idea of associative alignment for leveraging part of the base data by aligning the novel training instances to the closely related ones in the base training set. This expands the size of the effective novel training set by adding extra "related base" instances to the few novel ones, thereby allowing a constructive fine-tuning. We propose two associative alignment strategies: 1) a metric-learning loss for minimizing the distance between related base samples and the centroid of novel instances in the feature space, and 2) a conditional adversarial alignment loss based on the Wasserstein distance. Experiments on four standard datasets and three backbones demonstrate that combining our centroid-based alignment loss results in absolute accuracy improvements of 4.4%, 1.2%, and 6.2% in 5-shot learning over the state of the art for object recognition, fine-grained classification, and cross-domain adaptation, respectively.

Changjian Shui, Fan Zhou, Christian Gagné et al.

In this paper, we are proposing a unified and principled method for both the querying and training processes in deep batch active learning. We are providing theoretical insights from the intuition of modeling the interactive procedure in active learning as distribution matching, by adopting the Wasserstein distance. As a consequence, we derived a new training loss from the theoretical analysis, which is decomposed into optimizing deep neural network parameters and batch query selection through alternative optimization. In addition, the loss for training a deep neural network is naturally formulated as a min-max optimization problem through leveraging the unlabeled data information. Moreover, the proposed principles also indicate an explicit uncertainty-diversity trade-off in the query batch selection. Finally, we evaluate our proposed method on different benchmarks, consistently showing better empirical performances and a better time-efficient query strategy compared to the baselines.

Marc-André Gardner, Yannick Hold-Geoffroy, Kalyan Sunkavalli et al.

We present a method to estimate lighting from a single image of an indoor scene. Previous work has used an environment map representation that does not account for the localized nature of indoor lighting. Instead, we represent lighting as a set of discrete 3D lights with geometric and photometric parameters. We train a deep neural network to regress these parameters from a single image, on a dataset of environment maps annotated with depth. We propose a differentiable layer to convert these parameters to an environment map to compute our loss; this bypasses the challenge of establishing correspondences between estimated and ground truth lights. We demonstrate, via quantitative and qualitative evaluations, that our representation and training scheme lead to more accurate results compared to previous work, while allowing for more realistic 3D object compositing with spatially-varying lighting.

Azadeh Sadat Mozafari, Hugo Siqueira Gomes, Wilson Leão et al.

The great performances of deep learning are undeniable, with impressive results over a wide range of tasks. However, the output confidence of these models is usually not well-calibrated, which can be an issue for applications where confidence on the decisions is central to providing trust and reliability (e.g., autonomous driving or medical diagnosis). For models using softmax at the last layer, Temperature Scaling (TS) is a state-of-the-art calibration method, with low time and memory complexity as well as demonstrated effectiveness. TS relies on a T parameter to rescale and calibrate values of the softmax layer, whose parameter value is computed from a labelled dataset. We are proposing an Unsupervised Temperature Scaling (UTS) approach, which does not depend on labelled samples to calibrate the model, which allows, for example, the use of a part of a test samples to calibrate the pre-trained model before going into inference mode. We provide theoretical justifications for UTS and assess its effectiveness on a wide range of deep models and datasets. We also demonstrate calibration results of UTS on skin lesion detection, a problem where a well-calibrated output can play an important role for accurate decision-making.

Changjian Shui, Mahdieh Abbasi, Louis-Émile Robitaille et al.

Multitask learning aims at solving a set of related tasks simultaneously, by exploiting the shared knowledge for improving the performance on individual tasks. Hence, an important aspect of multitask learning is to understand the similarities within a set of tasks. Previous works have incorporated this similarity information explicitly (e.g., weighted loss for each task) or implicitly (e.g., adversarial loss for feature adaptation), for achieving good empirical performances. However, the theoretical motivations for adding task similarity knowledge are often missing or incomplete. In this paper, we give a different perspective from a theoretical point of view to understand this practice. We first provide an upper bound on the generalization error of multitask learning, showing the benefit of explicit and implicit task similarity knowledge. We systematically derive the bounds based on two distinct task similarity metrics: H divergence and Wasserstein distance. From these theoretical results, we revisit the Adversarial Multi-task Neural Network, proposing a new training algorithm to learn the task relation coefficients and neural network parameters iteratively. We assess our new algorithm empirically on several benchmarks, showing not only that we find interesting and robust task relations, but that the proposed approach outperforms the baselines, reaffirming the benefits of theoretical insight in algorithm design.

Sébastien de Blois, Ihsen Hedhli, Christian Gagné

To evaluate their performance, existing dehazing approaches generally rely on distance measures between the generated image and its corresponding ground truth. Despite its ability to produce visually good images, using pixel-based or even perceptual metrics do not guarantee, in general, that the produced image is fit for being used as input for low-level computer vision tasks such as segmentation. To overcome this weakness, we are proposing a novel end-to-end approach for image dehazing, fit for being used as input to an image segmentation procedure, while maintaining the visual quality of the generated images. Inspired by the success of Generative Adversarial Networks (GAN), we propose to optimize the generator by introducing a discriminator network and a loss function that evaluates segmentation quality of dehazed images. In addition, we make use of a supplementary loss function that verifies that the visual and the perceptual quality of the generated image are preserved in hazy conditions. Results obtained using the proposed technique are appealing, with a favorable comparison to state-of-the-art approaches when considering the performance of segmentation algorithms on the hazy images.

Azadeh Sadat Mozafari, Hugo Siqueira Gomes, Wilson Leão et al.

Recently, Deep Neural Networks (DNNs) have been achieving impressive results on wide range of tasks. However, they suffer from being well-calibrated. In decision-making applications, such as autonomous driving or medical diagnosing, the confidence of deep networks plays an important role to bring the trust and reliability to the system. To calibrate the deep networks' confidence, many probabilistic and measure-based approaches are proposed. Temperature Scaling (TS) is a state-of-the-art among measure-based calibration methods which has low time and memory complexity as well as effectiveness. In this paper, we study TS and show it does not work properly when the validation set that TS uses for calibration has small size or contains noisy-labeled samples. TS also cannot calibrate highly accurate networks as well as non-highly accurate ones. Accordingly, we propose Attended Temperature Scaling (ATS) which preserves the advantages of TS while improves calibration in aforementioned challenging situations. We provide theoretical justifications for ATS and assess its effectiveness on wide range of deep models and datasets. We also compare the calibration results of TS and ATS on skin lesion detection application as a practical problem where well-calibrated system can play important role in making a decision.

Changjian Shui, Ihsen Hedhli, Christian Gagné

Lifelong learning can be viewed as a continuous transfer learning procedure over consecutive tasks, where learning a given task depends on accumulated knowledge --- the so-called knowledge base. Most published work on lifelong learning makes a batch processing of each task, implying that a data collection step is required beforehand. We are proposing a new framework, lifelong online learning, in which the learning procedure for each task is interactive. This is done through a computationally efficient algorithm where the predicted result for a given task is made by combining two intermediate predictions: by using only the information from the current task and by relying on the accumulated knowledge. In this work, two challenges are tackled: making no assumption on the task generation distribution, and processing with a possibly unknown number of instances for each task. We are providing a theoretical analysis of this algorithm, with a cumulative error upper bound for each task. We find that under some mild conditions, the algorithm can still benefit from a small cumulative error even when facing few interactions. Moreover, we provide experimental results on both synthetic and real datasets that validate the correct behavior and practical usefulness of the proposed algorithm.

Alejandro Cervantes, Christian Gagné, Pedro Isasi et al.

Incremental learning from non-stationary data poses special challenges to the field of machine learning. Although new algorithms have been developed for this, assessment of results and comparison of behaviors are still open problems, mainly because evaluation metrics, adapted from more traditional tasks, can be ineffective in this context. Overall, there is a lack of common testing practices. This paper thus presents a testbed for incremental non-stationary learning algorithms, based on specially designed synthetic datasets. Also, test results are reported for some well-known algorithms to show that the proposed methodology is effective at characterizing their strengths and weaknesses. It is expected that this methodology will provide a common basis for evaluating future contributions in the field.

Mahdieh Abbasi, Arezoo Rajabi, Christian Gagné et al.

Detection and rejection of adversarial examples in security sensitive and safety-critical systems using deep CNNs is essential. In this paper, we propose an approach to augment CNNs with out-distribution learning in order to reduce misclassification rate by rejecting adversarial examples. We empirically show that our augmented CNNs can either reject or classify correctly most adversarial examples generated using well-known methods ( >95% for MNIST and >75% for CIFAR-10 on average). Furthermore, we achieve this without requiring to train using any specific type of adversarial examples and without sacrificing the accuracy of models on clean samples significantly (< 4%).

Louis-Émile Robitaille, Audrey Durand, Marc-André Gardner et al.

With super-resolution optical microscopy, it is now possible to observe molecular interactions in living cells. The obtained images have a very high spatial precision but their overall quality can vary a lot depending on the structure of interest and the imaging parameters. Moreover, evaluating this quality is often difficult for non-expert users. In this work, we tackle the problem of learning the quality function of super- resolution images from scores provided by experts. More specifically, we are proposing a system based on a deep neural network that can provide a quantitative quality measure of a STED image of neuronal structures given as input. We conduct a user study in order to evaluate the quality of the predictions of the neural network against those of a human expert. Results show the potential while highlighting some of the limits of the proposed approach.

Joel Lehman, Jeff Clune, Dusan Misevic et al.

Biological evolution provides a creative fount of complex and subtle adaptations, often surprising the scientists who discover them. However, because evolution is an algorithmic process that transcends the substrate in which it occurs, evolution's creativity is not limited to nature. Indeed, many researchers in the field of digital evolution have observed their evolving algorithms and organisms subverting their intentions, exposing unrecognized bugs in their code, producing unexpected adaptations, or exhibiting outcomes uncannily convergent with ones in nature. Such stories routinely reveal creativity by evolution in these digital worlds, but they rarely fit into the standard scientific narrative. Instead they are often treated as mere obstacles to be overcome, rather than results that warrant study in their own right. The stories themselves are traded among researchers through oral tradition, but that mode of information transmission is inefficient and prone to error and outright loss. Moreover, the fact that these stories tend to be shared only among practitioners means that many natural scientists do not realize how interesting and lifelike digital organisms are and how natural their evolution can be. To our knowledge, no collection of such anecdotes has been published before. This paper is the crowd-sourced product of researchers in the fields of artificial life and evolutionary computation who have provided first-hand accounts of such cases. It thus serves as a written, fact-checked collection of scientifically important and even entertaining stories. In doing so we also present here substantial evidence that the existence and importance of evolutionary surprises extends beyond the natural world, and may indeed be a universal property of all complex evolving systems.

Changjian Shui, Azadeh Sadat Mozafari, Jonathan Marek et al.

Calibrating the confidence of supervised learning models is important for a variety of contexts where the certainty over predictions should be reliable. However, it has been reported that deep neural network models are often too poorly calibrated for achieving complex tasks requiring reliable uncertainty estimates in their prediction. In this work, we are proposing a strategy for training deep ensembles with a diversity function regularization, which improves the calibration property while maintaining a similar prediction accuracy.

Mahdieh Abbasi, Christian Gagné

The easiness at which adversarial instances can be generated in deep neural networks raises some fundamental questions on their functioning and concerns on their use in critical systems. In this paper, we draw a connection between over-generalization and adversaries: a possible cause of adversaries lies in models designed to make decisions all over the input space, leading to inappropriate high-confidence decisions in parts of the input space not represented in the training set. We empirically show an augmented neural network, which is not trained on any types of adversaries, can increase the robustness by detecting black-box one-step adversaries, i.e. assimilated to out-distribution samples, and making generation of white-box one-step adversaries harder.

Marc-André Gardner, Kalyan Sunkavalli, Ersin Yumer et al.

We propose an automatic method to infer high dynamic range illumination from a single, limited field-of-view, low dynamic range photograph of an indoor scene. In contrast to previous work that relies on specialized image capture, user input, and/or simple scene models, we train an end-to-end deep neural network that directly regresses a limited field-of-view photo to HDR illumination, without strong assumptions on scene geometry, material properties, or lighting. We show that this can be accomplished in a three step process: 1) we train a robust lighting classifier to automatically annotate the location of light sources in a large dataset of LDR environment maps, 2) we use these annotations to train a deep neural network that predicts the location of lights in a scene from a single limited field-of-view photo, and 3) we fine-tune this network using a small dataset of HDR environment maps to predict light intensities. This allows us to automatically recover high-quality HDR illumination estimates that significantly outperform previous state-of-the-art methods. Consequently, using our illumination estimates for applications like 3D object insertion, we can achieve results that are photo-realistic, which is validated via a perceptual user study.

Mahdieh Abbasi, Christian Gagné

We are proposing to use an ensemble of diverse specialists, where speciality is defined according to the confusion matrix. Indeed, we observed that for adversarial instances originating from a given class, labeling tend to be done into a small subset of (incorrect) classes. Therefore, we argue that an ensemble of specialists should be better able to identify and reject fooling instances, with a high entropy (i.e., disagreement) over the decisions in the presence of adversaries. Experimental results obtained confirm that interpretation, opening a way to make the system more robust to adversarial examples through a rejection mechanism, rather than trying to classify them properly at any cost.

Audrey Durand, Christian Gagné

Many real-world applications are characterized by a number of conflicting performance measures. As optimizing in a multi-objective setting leads to a set of non-dominated solutions, a preference function is required for selecting the solution with the appropriate trade-off between the objectives. The question is: how good do estimations of these objectives have to be in order for the solution maximizing the preference function to remain unchanged? In this paper, we introduce the concept of preference radius to characterize the robustness of the preference function and provide guidelines for controlling the quality of estimations in the multi-objective setting. More specifically, we provide a general formulation of multi-objective optimization under the bandits setting. We show how the preference radius relates to the optimal gap and we use this concept to provide a theoretical analysis of the Thompson sampling algorithm from multivariate normal priors. We finally present experiments to support the theoretical results and highlight the fact that one cannot simply scalarize multi-objective problems into single-objective problems.

Farkhondeh Kiaee, Christian Gagné, Mahdieh Abbasi

The storage and computation requirements of Convolutional Neural Networks (CNNs) can be prohibitive for exploiting these models over low-power or embedded devices. This paper reduces the computational complexity of the CNNs by minimizing an objective function, including the recognition loss that is augmented with a sparsity-promoting penalty term. The sparsity structure of the network is identified using the Alternating Direction Method of Multipliers (ADMM), which is widely used in large optimization problems. This method alternates between promoting the sparsity of the network and optimizing the recognition performance, which allows us to exploit the two-part structure of the corresponding objective functions. In particular, we take advantage of the separability of the sparsity-inducing penalty functions to decompose the minimization problem into sub-problems that can be solved sequentially. Applying our method to a variety of state-of-the-art CNN models, our proposed method is able to simplify the original model, generating models with less computation and fewer parameters, while maintaining and often improving generalization performance. Accomplishments on a variety of models strongly verify that our proposed ADMM-based method can be a very useful tool for simplifying and improving deep CNNs.

Julien-Charles Lévesque, Christian Gagné, Robert Sabourin

In this paper, we bridge the gap between hyperparameter optimization and ensemble learning by performing Bayesian optimization of an ensemble with regards to its hyperparameters. Our method consists in building a fixed-size ensemble, optimizing the configuration of one classifier of the ensemble at each iteration of the hyperparameter optimization algorithm, taking into consideration the interaction with the other models when evaluating potential performances. We also consider the case where the ensemble is to be reconstructed at the end of the hyperparameter optimization phase, through a greedy selection over the pool of models generated during the optimization. We study the performance of our proposed method on three different hyperparameter spaces, showing that our approach is better than both the best single model and a greedy ensemble construction over the models produced by a standard Bayesian optimization.

François-Michel De Rainville, Michèle Sebag, Christian Gagné et al.

This paper proposes a self-adaptation mechanism to manage the resources allocated to the different species comprising a cooperative coevolutionary algorithm. The proposed approach relies on a dynamic extension to the well-known multi-armed bandit framework. At each iteration, the dynamic multi-armed bandit makes a decision on which species to evolve for a generation, using the history of progress made by the different species to guide the decisions. We show experimentally, on a benchmark and a real-world problem, that evolving the different populations at different paces allows not only to identify solutions more rapidly, but also improves the capacity of cooperative coevolution to solve more complex problems.