Yu Lu

Mavina Nikandrou, Lu Yu, Alessandro Suglia et al.

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.

Yu Lu, Junwei Bao, Zichen Ma et al.

High-quality data is essential for conversational recommendation systems and serves as the cornerstone of the network architecture development and training strategy design. Existing works contribute heavy human efforts to manually labeling or designing and extending recommender dialogue templates. However, they suffer from (i) the limited number of human annotators results in that datasets can hardly capture rich and large-scale cases in the real world, (ii) the limited experience and knowledge of annotators account for the uninformative corpus and inappropriate recommendations. In this paper, we propose a novel automatic dataset synthesis approach that can generate both large-scale and high-quality recommendation dialogues through a data2text generation process, where unstructured recommendation conversations are generated from structured graphs based on user-item information from the real world. In doing so, we comprehensively exploit: (i) rich personalized user profiles from traditional recommendation datasets, (ii) rich external knowledge from knowledge graphs, and (iii) the conversation ability contained in human-to-human conversational recommendation datasets. Extensive experiments validate the benefit brought by the automatically synthesized data under low-resource scenarios and demonstrate the promising potential to facilitate the development of a more effective conversational recommendation system.

Lu Yu, Wei Xiang

Recently vision transformer models have become prominent models for a range of tasks. These models, however, usually suffer from intensive computational costs and heavy memory requirements, making them impractical for deployment on edge platforms. Recent studies have proposed to prune transformers in an unexplainable manner, which overlook the relationship between internal units of the model and the target class, thereby leading to inferior performance. To alleviate this problem, we propose a novel explainable pruning framework dubbed X-Pruner, which is designed by considering the explainability of the pruning criterion. Specifically, to measure each prunable unit's contribution to predicting each target class, a novel explainability-aware mask is proposed and learned in an end-to-end manner. Then, to preserve the most informative units and learn the layer-wise pruning rate, we adaptively search the layer-wise threshold that differentiates between unpruned and pruned units based on their explainability-aware mask values. To verify and evaluate our method, we apply the X-Pruner on representative transformer models including the DeiT and Swin Transformer. Comprehensive simulation results demonstrate that the proposed X-Pruner outperforms the state-of-the-art black-box methods with significantly reduced computational costs and slight performance degradation.

Lu Yu, Wei Xiang, Juan Fang et al.

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.

Xingyu Liu, Alex Leonardi, Lu Yu et al.

In many practical scenarios -- like hyperparameter search or continual retraining with new data -- related training runs are performed many times in sequence. Current practice is to train each of these models independently from scratch. We study the problem of exploiting the computation invested in previous runs to reduce the cost of future runs using knowledge distillation (KD). We find that augmenting future runs with KD from previous runs dramatically reduces the time necessary to train these models, even taking into account the overhead of KD. We improve on these results with two strategies that reduce the overhead of KD by 80-90% with minimal effect on accuracy and vast pareto-improvements in overall cost. We conclude that KD is a promising avenue for reducing the cost of the expensive preparatory work that precedes training final models in practice.

Yu Lu, Roummel F. Marcia

The negative binomial model, which generalizes the Poisson distribution model, can be found in applications involving low-photon signal recovery, including medical imaging. Recent studies have explored several regularization terms for the negative binomial model, such as the $\ell_p$ quasi-norm with $0 < p < 1$, $\ell_1$ norm, and the total variation (TV) quasi-seminorm for promoting sparsity in signal recovery. These penalty terms have been shown to improve image reconstruction outcomes. In this paper, we investigate the $\ell_p$ quasi-seminorm, both isotropic and anisotropic $\ell_p$ TV quasi-seminorms, within the framework of the negative binomial statistical model. This problem can be formulated as an optimization problem, which we solve using a gradient-based approach. We present comparisons between the negative binomial and Poisson statistical models using the $\ell_p$ TV quasi-seminorm as well as common penalty terms. Our experimental results highlight the efficacy of the proposed method.

Yu Lu, Kevin Bui, Roummel F. Marcia

In many applications such as medical imaging, the measurement data represent counts of photons hitting a detector. Such counts in low-photon settings are often modeled using a Poisson distribution. However, this model assumes that the mean and variance of the signal's noise distribution are equal. For overdispersed data where the variance is greater than the mean, the negative binomial distribution is a more appropriate statistical model. In this paper, we propose an optimization approach for recovering images corrupted by overdispersed Poisson noise. In particular, we incorporate a weighted anisotropic-isotropic total variation regularizer, which avoids staircasing artifacts that are introduced by a regular total variation penalty. We use an alternating direction method of multipliers, where each subproblem has a closed-form solution. Numerical experiments demonstrate the effectiveness of our proposed approach, especially in very photon-limited settings.

Yifeng Yu, Lu Yu

Score-based diffusion models have emerged as powerful tools in generative modeling, yet their theoretical foundations remain underexplored. In this work, we focus on the Wasserstein convergence analysis of score-based diffusion models. Specifically, we investigate the impact of various discretization schemes, including Euler discretization, exponential integrators, and midpoint randomization methods. Our analysis provides a quantitative comparison of these discrete approximations, emphasizing their influence on convergence behavior. Furthermore, we explore scenarios where Hessian information is available and propose an accelerated sampler based on the local linearization method. We demonstrate that this Hessian-based approach achieves faster convergence rates of order $\widetilde{\mathcal{O}}\left(\frac{1}{\varepsilon}\right)$ significantly improving upon the standard rate $\widetilde{\mathcal{O}}\left(\frac{1}{\varepsilon^2}\right)$ of vanilla diffusion models, where $\varepsilon$ denotes the target accuracy.

Shuai Zhang, Yu Lu, Junwen Liu et al.

With the growing humanlike nature of dialog agents, people are now engaging in extended conversations that can stretch from brief moments to substantial periods of time. Understanding the factors that contribute to sustaining these interactions is crucial, yet existing studies primarily focusing on short-term simulations that rarely explore such prolonged and real conversations. In this paper, we investigate the factors influencing retention rates in real interactions with roleplaying models. By analyzing a large dataset of interactions between real users and thousands of characters, we systematically examine multiple factors and assess their impact on user retention rate. Surprisingly, we find that the degree to which the bot embodies the roles it plays has limited influence on retention rates, while the length of each turn it speaks significantly affects retention rates. This study sheds light on the critical aspects of user engagement with role-playing models and provides valuable insights for future improvements in the development of large language models for role-playing purposes.

Yukun Zuo, Hantao Yao, Lu Yu et al.

Continual learning endeavors to equip the model with the capability to integrate current task knowledge while mitigating the forgetting of past task knowledge. Inspired by prompt tuning, prompt-based methods maintain a frozen backbone and train with slight learnable prompts to minimize the catastrophic forgetting that arises due to updating a large number of backbone parameters. Nonetheless, these learnable prompts tend to concentrate on the discriminatory knowledge of the current task while ignoring past task knowledge, leading to that learnable prompts still suffering from catastrophic forgetting. This paper introduces a novel rehearsal-free paradigm for continual learning termed Hierarchical Prompts (H-Prompts), comprising three categories of prompts -- class prompt, task prompt, and general prompt. To effectively depict the knowledge of past classes, class prompt leverages Bayesian Distribution Alignment to model the distribution of classes in each task. To reduce the forgetting of past task knowledge, task prompt employs Cross-task Knowledge Excavation to amalgamate the knowledge encapsulated in the learned class prompts of past tasks and current task knowledge. Furthermore, general prompt utilizes Generalized Knowledge Exploration to deduce highly generalized knowledge in a self-supervised manner. Evaluations on two benchmarks substantiate the efficacy of the proposed H-Prompts, exemplified by an average accuracy of 87.8% in Split CIFAR-100 and 70.6% in Split ImageNet-R.

Nuri Mert Vural, Lu Yu, Krishnakumar Balasubramanian et al.

We study stochastic convex optimization under infinite noise variance. Specifically, when the stochastic gradient is unbiased and has uniformly bounded $(1+κ)$-th moment, for some $κ\in (0,1]$, 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.

Alex Gomez-Villa, Bartlomiej Twardowski, Lu Yu et al.

Recent self-supervised learning methods are able to learn high-quality image representations and are closing the gap with supervised approaches. However, these methods are unable to acquire new knowledge incrementally -- they are, in fact, mostly used only as a pre-training phase over IID data. In this work we investigate self-supervised methods in continual learning regimes without any replay mechanism. We show that naive functional regularization, also known as feature distillation, leads to lower plasticity and limits continual learning performance. Instead, we propose Projected Functional Regularization in which a separate temporal projection network ensures that the newly learned feature space preserves information of the previous one, while at the same time allowing for the learning of new features. This prevents forgetting while maintaining the plasticity of the learner. Comparison with other incremental learning approaches applied to self-supervision demonstrates that our method obtains competitive performance in different scenarios and on multiple datasets.

Lu Yu, Xialei Liu, Joost van de Weijer

In class-incremental semantic segmentation, we have no access to the labeled data of previous tasks. Therefore, when incrementally learning new classes, deep neural networks suffer from catastrophic forgetting of previously learned knowledge. To address this problem, we propose to apply a self-training approach that leverages unlabeled data, which is used for the rehearsal of previous knowledge. Specifically, we first learn a temporary model for the current task, and then pseudo labels for the unlabeled data are computed by fusing information from the old model of the previous task and the current temporary model. Additionally, conflict reduction is proposed to resolve the conflicts of pseudo labels generated from both the old and temporary models. We show that maximizing self-entropy can further improve results by smoothing the overconfident predictions. Interestingly, in the experiments we show that the auxiliary data can be different from the training data and that even general-purpose but diverse auxiliary data can lead to large performance gains. The experiments demonstrate state-of-the-art results: obtaining a relative gain of up to 114% on Pascal-VOC 2012 and 8.5% on the more challenging ADE20K compared to previous state-of-the-art methods.

Lu Yu, Shichao Pei, Lizhong Ding et al.

This paper studies learning node representations with graph neural networks (GNNs) for unsupervised scenario. Specifically, we derive a theoretical analysis and provide an empirical demonstration about the non-steady performance of GNNs over different graph datasets, when the supervision signals are not appropriately defined. The performance of GNNs depends on both the node feature smoothness and the locality of graph structure. To smooth the discrepancy of node proximity measured by graph topology and node feature, we proposed SAIL - a novel \underline{S}elf-\underline{A}ugmented graph contrast\underline{i}ve \underline{L}earning framework, with two complementary self-distilling regularization modules, \emph{i.e.}, intra- and inter-graph knowledge distillation. We demonstrate the competitive performance of SAIL on a variety of graph applications. Even with a single GNN layer, SAIL has consistently competitive or even better performance on various benchmark datasets, comparing with state-of-the-art baselines.

Johannes Lederer, Lu Yu, Irina Gaynanova

The abundance of high-dimensional data in the modern sciences has generated tremendous interest in penalized estimators such as the lasso, scaled lasso, square-root lasso, elastic net, and many others. In this paper, we establish a general oracle inequality for prediction in high-dimensional linear regression with such methods. Since the proof relies only on convexity and continuity arguments, the result holds irrespective of the design matrix and applies to a wide range of penalized estimators. Overall, the bound demonstrates that generic estimators can provide consistent prediction with any design matrix. From a practical point of view, the bound can help to identify the potential of specific estimators, and they can help to get a sense of the prediction accuracy in a given application.

Chao Gao, Yu Lu, Zongming Ma et al.

Biclustering structures in data matrices were first formalized in a seminal paper by John Hartigan (1972) where one seeks to cluster cases and variables simultaneously. Such structures are also prevalent in block modeling of networks. In this paper, we develop a unified theory for the estimation and completion of matrices with biclustering structures, where the data is a partially observed and noise contaminated data matrix with a certain biclustering structure. In particular, we show that a constrained least squares estimator achieves minimax rate-optimal performance in several of the most important scenarios. To this end, we derive unified high probability upper bounds for all sub-Gaussian data and also provide matching minimax lower bounds in both Gaussian and binary cases. Due to the close connection of graphon to stochastic block models, an immediate consequence of our general results is a minimax rate-optimal estimator for sparse graphons.

Yu Lu, Sahand N. Negahban

In recent years rank aggregation has received significant attention from the machine learning community. The goal of such a problem is to combine the (partially revealed) preferences over objects of a large population into a single, relatively consistent ordering of those objects. However, in many cases, we might not want a single ranking and instead opt for individual rankings. We study a version of the problem known as collaborative ranking. In this problem we assume that individual users provide us with pairwise preferences (for example purchasing one item over another). From those preferences we wish to obtain rankings on items that the users have not had an opportunity to explore. The results here have a very interesting connection to the standard matrix completion problem. We provide a theoretical justification for a nuclear norm regularized optimization procedure, and provide high-dimensional scaling results that show how the error in estimating user preferences behaves as the number of observations increase.