Continual learning aims to train a model incrementally on a sequence of tasks without forgetting previous knowledge. Although continual learning has been widely studied in computer vision, its application to Vision+Language tasks is not that straightforward, as settings can be parameterized in multiple ways according to their input modalities. In this paper, we present a detailed study of how different settings affect performance for Visual Question Answering. We first propose three plausible task formulations and demonstrate their impact on the performance of continual learning algorithms. We break down several factors of task similarity, showing that performance and sensitivity to task order highly depend on the shift of the output distribution. We also investigate the potential of pretrained models and compare the robustness of transformer models with different visual embeddings. Finally, we provide an analysis interpreting model representations and their impact on forgetting. Our results highlight the importance of stabilizing visual representations in deeper layers.
The cold-start problem has been commonly recognized in recommendation systems and studied by following a general idea to leverage the abundant interaction records of warm users to infer the preference of cold users. However, the performance of these solutions is limited by the amount of records available from warm users to use. Thus, building a recommendation system based on few interaction records from a few users still remains a challenging problem for unpopular or early-stage recommendation platforms. This paper focuses on solving the few-shot recommendation problem for news recommendation based on two observations. First, news at different platforms (even in different languages) may share similar topics. Second, the user preference over these topics is transferable across different platforms. Therefore, we propose to solve the few-shot news recommendation problem by transferring the user-news preference from a rich source domain to a low-resource target domain. To bridge two domains in different languages without any overlapping users and news, we propose a novel unsupervised cross-lingual transfer model as the news encoder that aligns semantically similar news in two domains. A user encoder is constructed on top of the aligned news encoding and transfers the user preference from the source to the target domain. Experimental results on two real-world news recommendation datasets show the superior performance of our proposed method on addressing few-shot news recommendation, comparing to the state-of-the-art baselines.
Recently vision transformer models have become prominent models for a range of vision tasks. These models, however, are usually opaque with weak feature interpretability. Moreover, there is no method currently built for an intrinsically interpretable transformer, which is able to explain its reasoning process and provide a faithful explanation. To close these crucial gaps, we propose a novel vision transformer dubbed the eXplainable Vision Transformer (eX-ViT), an intrinsically interpretable transformer model that is able to jointly discover robust interpretable features and perform the prediction. Specifically, eX-ViT is composed of the Explainable Multi-Head Attention (E-MHA) module, the Attribute-guided Explainer (AttE) module and the self-supervised attribute-guided loss. The E-MHA tailors explainable attention weights that are able to learn semantically interpretable representations from local patches in terms of model decisions with noise robustness. Meanwhile, AttE is proposed to encode discriminative attribute features for the target object through diverse attribute discovery, which constitutes faithful evidence for the model's predictions. In addition, a self-supervised attribute-guided loss is developed for our eX-ViT, which aims at learning enhanced representations through the attribute discriminability mechanism and attribute diversity mechanism, to localize diverse and discriminative attributes and generate more robust explanations. As a result, we can uncover faithful and robust interpretations with diverse attributes through the proposed eX-ViT.
Adversarial robustness evaluates the worst-case performance scenario of a machine learning model to ensure its safety and reliability. This study is the first to investigate the robustness of visually grounded dialog models towards textual attacks. These attacks represent a worst-case scenario where the input question contains a synonym which causes the previously correct model to return a wrong answer. Using this scenario, we first aim to understand how multimodal input components contribute to model robustness. Our results show that models which encode dialog history are more robust, and when launching an attack on history, model prediction becomes more uncertain. This is in contrast to prior work which finds that dialog history is negligible for model performance on this task. We also evaluate how to generate adversarial test examples which successfully fool the model but remain undetected by the user/software designer. We find that the textual, as well as the visual context are important to generate plausible worst-case scenarios.
Learned image compression (LIC) has reached a comparable coding gain with traditional hand-crafted methods such as VVC intra. However, the large network complexity prohibits the usage of LIC on resource-limited embedded systems. Network quantization is an efficient way to reduce the network burden. This paper presents a quantized LIC (QLIC) by channel splitting. First, we explore that the influence of quantization error to the reconstruction error is different for various channels. Second, we split the channels whose quantization has larger influence to the reconstruction error. After the splitting, the dynamic range of channels is reduced so that the quantization error can be reduced. Finally, we prune several channels to keep the number of overall channels as origin. By using the proposal, in the case of 8-bit quantization for weight and activation of both main and hyper path, we can reduce the BD-rate by 0.61%-4.74% compared with the previous QLIC. Besides, we can reach better coding gain compared with the state-of-the-art network quantization method when quantizing MS-SSIM models. Moreover, our proposal can be combined with other network quantization methods to further improve the coding gain. The moderate coding loss caused by the quantization validates the feasibility of the hardware implementation for QLIC in the future.
We study stochastic convex optimization under infinite noise variance. Specifically, when the stochastic gradient is unbiased and has uniformly bounded $(1+\kappa)$-th moment, for some $\kappa \in (0,1]$, we quantify the convergence rate of the Stochastic Mirror Descent algorithm with a particular class of uniformly convex mirror maps, in terms of the number of iterations, dimensionality and related geometric parameters of the optimization problem. Interestingly this algorithm does not require any explicit gradient clipping or normalization, which have been extensively used in several recent empirical and theoretical works. We complement our convergence results with information-theoretic lower bounds showing that no other algorithm using only stochastic first-order oracles can achieve improved rates. Our results have several interesting consequences for devising online/streaming stochastic approximation algorithms for problems arising in robust statistics and machine learning.
Consider a panel data setting where repeated observations on individuals are available. Often it is reasonable to assume that there exist groups of individuals that share similar effects of observed characteristics, but the grouping is typically unknown in advance. We propose a novel approach to estimate such unobserved groupings for general panel data models. Our method explicitly accounts for the uncertainty in individual parameter estimates and remains computationally feasible with a large number of individuals and/or repeated measurements on each individual. The developed ideas can be applied even when individual-level data are not available and only parameter estimates together with some quantification of uncertainty are given to the researcher.
Recent self-supervised learning methods are able to learn high-quality image representations and are closing the gap with supervised methods. However, these methods are unable to acquire new knowledge incrementally -- they are, in fact, mostly used only as a pre-training phase with IID data. In this work we investigate self-supervised methods in continual learning regimes without additional memory or replay. To prevent forgetting of previous knowledge, we propose the usage of functional regularization. We will show that naive functional regularization, also known as feature distillation, leads to low plasticity and therefore seriously limits continual learning performance. To address this problem, we propose Projected Functional Regularization where a separate projection network ensures that the newly learned feature space preserves information of the previous feature space, while allowing for the learning of new features. This allows us to prevent forgetting while maintaining the plasticity of the learner. Evaluation against other incremental learning approaches applied to self-supervision demonstrates that our method obtains competitive performance in different scenarios and on multiple datasets.
End-to-end Learned image compression (LIC) has reached the traditional hand-crafted methods such as BPG (HEVC intra) in terms of the coding gain. However, the large network size prohibits the usage of LIC on resource-limited embedded systems. This paper reduces the network complexity by quantizing both weights and activations. 1) For the weight quantization, we study different kinds of grouping and quantization scheme at first. A channel-wise non-linear quantization scheme is determined based on the coding gain analysis. After that, we propose a fine tuning scheme to clip the weights within a certain range so that the quantization error can be reduced. 2) For the activation quantization, we first propose multiple non-linear quantization codebooks with different maximum dynamic ranges. By selecting an optimal one through a multiplexer, the quantization range can be saturated to the greatest extent. In addition, we also exploit the mean-removed quantization for the analysis transform outputs in order to reduce the bit-width cost for the specific channel with the large non-zero mean. By quantizing each weight and activation element from 32-bit floating point to 8-bit fixed point, the memory cost for both weight and activation can be reduced by 75% with negligible coding performance loss. As a result, our quantized LIC can still outperform BPG in terms of MS-SSIM. To our best knowledge, this is the first work to give a complete analysis on the coding gain and the memory cost for a quantized LIC network, which validates the feasibility of the hardware implementation.
Intra prediction is an essential component in the image coding. This paper gives an intra prediction framework completely based on neural network modes (NM). Each NM can be regarded as a regression from the neighboring reference blocks to the current coding block. (1) For variable block size, we utilize different network structures. For small blocks 4x4 and 8x8, fully connected networks are used, while for large blocks 16x16 and 32x32, convolutional neural networks are exploited. (2) For each prediction mode, we develop a specific pre-trained network to boost the regression accuracy. When integrating into HEVC test model, we can save 3.55%, 3.03% and 3.27% BD-rate for Y, U, V components compared with the anchor. As far as we know, this is the first work to explore a fully NM based framework for intra prediction, and we reach a better coding gain with a lower complexity compared with the previous work.