Learning to Retain while Acquiring: Combating Distribution-Shift in Adversarial Data-Free Knowledge Distillation

Data-free Knowledge Distillation (DFKD) has gained popularity recently, with the fundamental idea of carrying out knowledge transfer from a Teacher neural network to a Student neural network in the absence of training data. However, in the Adversarial DFKD framework, the student network's accuracy, suffers due to the non-stationary distribution of the pseudo-samples under multiple generator updates. To this end, at every generator update, we aim to maintain the student's performance on previously encountered examples while acquiring knowledge from samples of the current distribution. Thus, we propose a meta-learning inspired framework by treating the task of Knowledge-Acquisition (learning from newly generated samples) and Knowledge-Retention (retaining knowledge on previously met samples) as meta-train and meta-test, respectively. Hence, we dub our method as Learning to Retain while Acquiring. Moreover, we identify an implicit aligning factor between the Knowledge-Retention and Knowledge-Acquisition tasks indicating that the proposed student update strategy enforces a common gradient direction for both tasks, alleviating interference between the two objectives. Finally, we support our hypothesis by exhibiting extensive evaluation and comparison of our method with prior arts on multiple datasets.

Energy-Inspired Self-Supervised Pretraining for Vision Models

Motivated by the fact that forward and backward passes of a deep network naturally form symmetric mappings between input and output representations, we introduce a simple yet effective self-supervised vision model pretraining framework inspired by energy-based models (EBMs). In the proposed framework, we model energy estimation and data restoration as the forward and backward passes of a single network without any auxiliary components, e.g., an extra decoder. For the forward pass, we fit a network to an energy function that assigns low energy scores to samples that belong to an unlabeled dataset, and high energy otherwise. For the backward pass, we restore data from corrupted versions iteratively using gradient-based optimization along the direction of energy minimization. In this way, we naturally fold the encoder-decoder architecture widely used in masked image modeling into the forward and backward passes of a single vision model. Thus, our framework now accepts a wide range of pretext tasks with different data corruption methods, and permits models to be pretrained from masked image modeling, patch sorting, and image restoration, including super-resolution, denoising, and colorization. We support our findings with extensive experiments, and show the proposed method delivers comparable and even better performance with remarkably fewer epochs of training compared to the state-of-the-art self-supervised vision model pretraining methods. Our findings shed light on further exploring self-supervised vision model pretraining and pretext tasks beyond masked image modeling.

Recycling Model Updates in Federated Learning: Are Gradient Subspaces Low-Rank?

In this paper, we question the rationale behind propagating large numbers of parameters through a distributed system during federated learning. We start by examining the rank characteristics of the subspace spanned by gradients across epochs (i.e., the gradient-space) in centralized model training, and observe that this gradient-space often consists of a few leading principal components accounting for an overwhelming majority (95-99%) of the explained variance. Motivated by this, we propose the "Look-back Gradient Multiplier" (LBGM) algorithm, which exploits this low-rank property to enable gradient recycling between model update rounds of federated learning, reducing transmissions of large parameters to single scalars for aggregation. We analytically characterize the convergence behavior of LBGM, revealing the nature of the trade-off between communication savings and model performance. Our subsequent experimental results demonstrate the improvement LBGM obtains in communication overhead compared to conventional federated learning on several datasets and deep learning models. Additionally, we show that LBGM is a general plug-and-play algorithm that can be used standalone or stacked on top of existing sparsification techniques for distributed model training.

Exploiting a Zoo of Checkpoints for Unseen Tasks

There are so many models in the literature that it is difficult for practitioners to decide which combinations are likely to be effective for a new task. This paper attempts to address this question by capturing relationships among checkpoints published on the web. We model the space of tasks as a Gaussian process. The covariance can be estimated from checkpoints and unlabeled probing data. With the Gaussian process, we can identify representative checkpoints by a maximum mutual information criterion. This objective is submodular. A greedy method identifies representatives that are likely to "cover" the task space. These representatives generalize to new tasks with superior performance. Empirical evidence is provided for applications from both computational linguistics as well as computer vision.

TESSERACT: Gradient Flip Score to Secure Federated Learning Against Model Poisoning Attacks

Federated learning---multi-party, distributed learning in a decentralized environment---is vulnerable to model poisoning attacks, even more so than centralized learning approaches. This is because malicious clients can collude and send in carefully tailored model updates to make the global model inaccurate. This motivated the development of Byzantine-resilient federated learning algorithms, such as Krum, Bulyan, FABA, and FoolsGold. However, a recently developed untargeted model poisoning attack showed that all prior defenses can be bypassed. The attack uses the intuition that simply by changing the sign of the gradient updates that the optimizer is computing, for a set of malicious clients, a model can be diverted from the optima to increase the test error rate. In this work, we develop TESSERACT---a defense against this directed deviation attack, a state-of-the-art model poisoning attack. TESSERACT is based on a simple intuition that in a federated learning setting, certain patterns of gradient flips are indicative of an attack. This intuition is remarkably stable across different learning algorithms, models, and datasets. TESSERACT assigns reputation scores to the participating clients based on their behavior during the training phase and then takes a weighted contribution of the clients. We show that TESSERACT provides robustness against even a white-box version of the attack.

Adaptive Convolutions with Per-pixel Dynamic Filter Atom

Applying feature dependent network weights have been proved to be effective in many fields. However, in practice, restricted by the enormous size of model parameters and memory footprints, scalable and versatile dynamic convolutions with per-pixel adapted filters are yet to be fully explored. In this paper, we address this challenge by decomposing filters, adapted to each spatial position, over dynamic filter atoms generated by a light-weight network from local features. Adaptive receptive fields can be supported by further representing each filter atom over sets of pre-fixed multi-scale bases. As plug-and-play replacements to convolutional layers, the introduced adaptive convolutions with per-pixel dynamic atoms enable explicit modeling of intra-image variance, while avoiding heavy computation, parameters, and memory cost. Our method preserves the appealing properties of conventional convolutions as being translation-equivariant and parametrically efficient. We present experiments to show that, the proposed method delivers comparable or even better performance across tasks, and are particularly effective on handling tasks with significant intra-image variance.

Cirrus: A Long-range Bi-pattern LiDAR Dataset

In this paper, we introduce Cirrus, a new long-range bi-pattern LiDAR public dataset for autonomous driving tasks such as 3D object detection, critical to highway driving and timely decision making. Our platform is equipped with a high-resolution video camera and a pair of LiDAR sensors with a 250-meter effective range, which is significantly longer than existing public datasets. We record paired point clouds simultaneously using both Gaussian and uniform scanning patterns. Point density varies significantly across such a long range, and different scanning patterns further diversify object representation in LiDAR. In Cirrus, eight categories of objects are exhaustively annotated in the LiDAR point clouds for the entire effective range. To illustrate the kind of studies supported by this new dataset, we introduce LiDAR model adaptation across different ranges, scanning patterns, and sensor devices. Promising results show the great potential of this new dataset to the robotics and computer vision communities.

Using Text to Teach Image Retrieval

Image retrieval relies heavily on the quality of the data modeling and the distance measurement in the feature space. Building on the concept of image manifold, we first propose to represent the feature space of images, learned via neural networks, as a graph. Neighborhoods in the feature space are now defined by the geodesic distance between images, represented as graph vertices or manifold samples. When limited images are available, this manifold is sparsely sampled, making the geodesic computation and the corresponding retrieval harder. To address this, we augment the manifold samples with geometrically aligned text, thereby using a plethora of sentences to teach us about images. In addition to extensive results on standard datasets illustrating the power of text to help in image retrieval, a new public dataset based on CLEVR is introduced to quantify the semantic similarity between visual data and text data. The experimental results show that the joint embedding manifold is a robust representation, allowing it to be a better basis to perform image retrieval given only an image and a textual instruction on the desired modifications over the image

Learning to Learn Variational Semantic Memory

In this paper, we introduce variational semantic memory into meta-learning to acquire long-term knowledge for few-shot learning. The variational semantic memory accrues and stores semantic information for the probabilistic inference of class prototypes in a hierarchical Bayesian framework. The semantic memory is grown from scratch and gradually consolidated by absorbing information from tasks it experiences. By doing so, it is able to accumulate long-term, general knowledge that enables it to learn new concepts of objects. We formulate memory recall as the variational inference of a latent memory variable from addressed contents, which offers a principled way to adapt the knowledge to individual tasks. Our variational semantic memory, as a new long-term memory module, confers principled recall and update mechanisms that enable semantic information to be efficiently accrued and adapted for few-shot learning. Experiments demonstrate that the probabilistic modelling of prototypes achieves a more informative representation of object classes compared to deterministic vectors. The consistent new state-of-the-art performance on four benchmarks shows the benefit of variational semantic memory in boosting few-shot recognition.

ACDC: Weight Sharing in Atom-Coefficient Decomposed Convolution

Convolutional Neural Networks (CNNs) are known to be significantly over-parametrized, and difficult to interpret, train and adapt. In this paper, we introduce a structural regularization across convolutional kernels in a CNN. In our approach, each convolution kernel is first decomposed as 2D dictionary atoms linearly combined by coefficients. The widely observed correlation and redundancy in a CNN hint a common low-rank structure among the decomposed coefficients, which is here further supported by our empirical observations. We then explicitly regularize CNN kernels by enforcing decomposed coefficients to be shared across sub-structures, while leaving each sub-structure only its own dictionary atoms, a few hundreds of parameters typically, which leads to dramatic model reductions. We explore models with sharing across different sub-structures to cover a wide range of trade-offs between parameter reduction and expressiveness. Our proposed regularized network structures open the door to better interpreting, training and adapting deep models. We validate the flexibility and compatibility of our method by image classification experiments on multiple datasets and underlying network structures, and show that CNNs now maintain performance with dramatic reduction in parameters and computations, e.g., only 5\% parameters are used in a ResNet-18 to achieve comparable performance. Further experiments on few-shot classification show that faster and more robust task adaptation is obtained in comparison with models with standard convolutions.