Relational Diffusion Distillation for Efficient Image Generation

Although the diffusion model has achieved remarkable performance in the field of image generation, its high inference delay hinders its wide application in edge devices with scarce computing resources. Therefore, many training-free sampling methods have been proposed to reduce the number of sampling steps required for diffusion models. However, they perform poorly under a very small number of sampling steps. Thanks to the emergence of knowledge distillation technology, the existing training scheme methods have achieved excellent results at very low step numbers. However, the current methods mainly focus on designing novel diffusion model sampling methods with knowledge distillation. How to transfer better diffusion knowledge from teacher models is a more valuable problem but rarely studied. Therefore, we propose Relational Diffusion Distillation (RDD), a novel distillation method tailored specifically for distilling diffusion models. Unlike existing methods that simply align teacher and student models at pixel level or feature distributions, our method introduces cross-sample relationship interaction during the distillation process and alleviates the memory constraints induced by multiple sample interactions. Our RDD significantly enhances the effectiveness of the progressive distillation framework within the diffusion model. Extensive experiments on several datasets (e.g., CIFAR-10 and ImageNet) demonstrate that our proposed RDD leads to 1.47 FID decrease under 1 sampling step compared to state-of-the-art diffusion distillation methods and achieving 256x speed-up compared to DDIM strategy. Code is available at https://github.com/cantbebetter2/RDD.

Continual Learning in the Frequency Domain

Continual learning (CL) is designed to learn new tasks while preserving existing knowledge. Replaying samples from earlier tasks has proven to be an effective method to mitigate the forgetting of previously acquired knowledge. However, the current research on the training efficiency of rehearsal-based methods is insufficient, which limits the practical application of CL systems in resource-limited scenarios. The human visual system (HVS) exhibits varying sensitivities to different frequency components, enabling the efficient elimination of visually redundant information. Inspired by HVS, we propose a novel framework called Continual Learning in the Frequency Domain (CLFD). To our knowledge, this is the first study to utilize frequency domain features to enhance the performance and efficiency of CL training on edge devices. For the input features of the feature extractor, CLFD employs wavelet transform to map the original input image into the frequency domain, thereby effectively reducing the size of input feature maps. Regarding the output features of the feature extractor, CLFD selectively utilizes output features for distinct classes for classification, thereby balancing the reusability and interference of output features based on the frequency domain similarity of the classes across various tasks. Optimizing only the input and output features of the feature extractor allows for seamless integration of CLFD with various rehearsal-based methods. Extensive experiments conducted in both cloud and edge environments demonstrate that CLFD consistently improves the performance of state-of-the-art (SOTA) methods in both precision and training efficiency. Specifically, CLFD can increase the accuracy of the SOTA CL method by up to 6.83% and reduce the training time by 2.6$\times$.

LightWeather: Harnessing Absolute Positional Encoding to Efficient and Scalable Global Weather Forecasting

Recently, Transformers have gained traction in weather forecasting for their capability to capture long-term spatial-temporal correlations. However, their complex architectures result in large parameter counts and extended training times, limiting their practical application and scalability to global-scale forecasting. This paper aims to explore the key factor for accurate weather forecasting and design more efficient solutions. Interestingly, our empirical findings reveal that absolute positional encoding is what really works in Transformer-based weather forecasting models, which can explicitly model the spatial-temporal correlations even without attention mechanisms. We theoretically prove that its effectiveness stems from the integration of geographical coordinates and real-world time features, which are intrinsically related to the dynamics of weather. Based on this, we propose LightWeather, a lightweight and effective model for station-based global weather forecasting. We employ absolute positional encoding and a simple MLP in place of other components of Transformer. With under 30k parameters and less than one hour of training time, LightWeather achieves state-of-the-art performance on global weather datasets compared to other advanced DL methods. The results underscore the superiority of integrating spatial-temporal knowledge over complex architectures, providing novel insights for DL in weather forecasting.

Unified Dual-Intent Translation for Joint Modeling of Search and Recommendation

Recommendation systems, which assist users in discovering their preferred items among numerous options, have served billions of users across various online platforms. Intuitively, users' interactions with items are highly driven by their unchanging inherent intents (e.g., always preferring high-quality items) and changing demand intents (e.g., wanting a T-shirt in summer but a down jacket in winter). However, both types of intents are implicitly expressed in recommendation scenario, posing challenges in leveraging them for accurate intent-aware recommendations. Fortunately, in search scenario, often found alongside recommendation on the same online platform, users express their demand intents explicitly through their query words. Intuitively, in both scenarios, a user shares the same inherent intent and the interactions may be influenced by the same demand intent. It is therefore feasible to utilize the interaction data from both scenarios to reinforce the dual intents for joint intent-aware modeling. But the joint modeling should deal with two problems: 1) accurately modeling users' implicit demand intents in recommendation; 2) modeling the relation between the dual intents and the interactive items. To address these problems, we propose a novel model named Unified Dual-Intents Translation for joint modeling of Search and Recommendation (UDITSR). To accurately simulate users' demand intents in recommendation, we utilize real queries from search data as supervision information to guide its generation. To explicitly model the relation among the triplet <inherent intent, demand intent, interactive item>, we propose a dual-intent translation propagation mechanism to learn the triplet in the same semantic space via embedding translations. Extensive experiments demonstrate that UDITSR outperforms SOTA baselines both in search and recommendation tasks.

Online Policy Distillation with Decision-Attention

Policy Distillation (PD) has become an effective method to improve deep reinforcement learning tasks. The core idea of PD is to distill policy knowledge from a teacher agent to a student agent. However, the teacher-student framework requires a well-trained teacher model which is computationally expensive.In the light of online knowledge distillation, we study the knowledge transfer between different policies that can learn diverse knowledge from the same environment.In this work, we propose Online Policy Distillation (OPD) with Decision-Attention (DA), an online learning framework in which different policies operate in the same environment to learn different perspectives of the environment and transfer knowledge to each other to obtain better performance together. With the absence of a well-performance teacher policy, the group-derived targets play a key role in transferring group knowledge to each student policy. However, naive aggregation functions tend to cause student policies quickly homogenize. To address the challenge, we introduce the Decision-Attention module to the online policies distillation framework. The Decision-Attention module can generate a distinct set of weights for each policy to measure the importance of group members. We use the Atari platform for experiments with various reinforcement learning algorithms, including PPO and DQN. In different tasks, our method can perform better than an independent training policy on both PPO and DQN algorithms. This suggests that our OPD-DA can transfer knowledge between different policies well and help agents obtain more rewards.

EGOR: Efficient Generated Objects Replay for incremental object detection

Incremental object detection aims to simultaneously maintain old-class accuracy and detect emerging new-class objects in incremental data. Most existing distillation-based methods underperform when unlabeled old-class objects are absent in the incremental dataset. While the absence can be mitigated by generating old-class samples, it also incurs high computational costs. In this paper, we argue that the extra computational cost stems from the inconsistency between the detector and the generative model, along with redundant generation. To overcome this problem, we propose Efficient Generated Object Replay (EGOR). Specifically, we generate old-class samples by inversing the original detectors, thus eliminating the necessity of training and storing additional generative models. We also propose augmented replay to reuse the objects in generated samples, thereby reducing the redundant generation. In addition, we propose high-response knowledge distillation focusing on the knowledge related to the old class, which transfers the knowledge in generated objects to the incremental detector. With the addition of the generated objects and losses, we observe a bias towards old classes in the detector. We balance the losses for old and new classes to alleviate the bias, thereby increasing the overall detection accuracy. Extensive experiments conducted on MS COCO 2017 demonstrate that our method can efficiently improve detection performance in the absence of old-class objects.

Exemplar-Free Class Incremental Learning via Incremental Representation

Exemplar-Free Class Incremental Learning (efCIL) aims to continuously incorporate the knowledge from new classes while retaining previously learned information, without storing any old-class exemplars (i.e., samples). For this purpose, various efCIL methods have been proposed over the past few years, generally with elaborately constructed old pseudo-features, increasing the difficulty of model development and interpretation. In contrast, we propose a \textbf{simple Incremental Representation (IR) framework} for efCIL without constructing old pseudo-features. IR utilizes dataset augmentation to cover a suitable feature space and prevents the model from forgetting by using a single L2 space maintenance loss. We discard the transient classifier trained on each one of the sequence tasks and instead replace it with a 1-near-neighbor classifier for inference, ensuring the representation is incrementally updated during CIL. Extensive experiments demonstrate that our proposed IR achieves comparable performance while significantly preventing the model from forgetting on CIFAR100, TinyImageNet, and ImageNetSubset datasets.

E2Net: Resource-Efficient Continual Learning with Elastic Expansion Network

Continual Learning methods are designed to learn new tasks without erasing previous knowledge. However, Continual Learning often requires massive computational power and storage capacity for satisfactory performance. In this paper, we propose a resource-efficient continual learning method called the Elastic Expansion Network (E2Net). Leveraging core subnet distillation and precise replay sample selection, E2Net achieves superior average accuracy and diminished forgetting within the same computational and storage constraints, all while minimizing processing time. In E2Net, we propose Representative Network Distillation to identify the representative core subnet by assessing parameter quantity and output similarity with the working network, distilling analogous subnets within the working network to mitigate reliance on rehearsal buffers and facilitating knowledge transfer across previous tasks. To enhance storage resource utilization, we then propose Subnet Constraint Experience Replay to optimize rehearsal efficiency through a sample storage strategy based on the structures of representative networks. Extensive experiments conducted predominantly on cloud environments with diverse datasets and also spanning the edge environment demonstrate that E2Net consistently outperforms state-of-the-art methods. In addition, our method outperforms competitors in terms of both storage and computational requirements.

CLIP-KD: An Empirical Study of Distilling CLIP Models

CLIP has become a promising language-supervised visual pre-training framework and achieves excellent performance over a wide range of tasks. This paper aims to distill small CLIP models supervised by a large teacher CLIP model. We propose several distillation strategies, including relation, feature, gradient and contrastive paradigm, to examine the impact on CLIP distillation. We show that the simplest feature mimicry with MSE loss performs best. Moreover, interactive contrastive learning and relation-based distillation are also critical in performance improvement. We apply the unified method to distill several student networks trained on 15 million (image, text) pairs. Distillation improves the student CLIP models consistently over zero-shot ImageNet classification and cross-modal retrieval benchmarks. We hope our empirical study will become an important baseline for future CLIP distillation research. The code is available at \url{https://github.com/winycg/CLIP-KD}.

Categories of Response-Based, Feature-Based, and Relation-Based Knowledge Distillation

Deep neural networks have achieved remarkable performance for artificial intelligence tasks. The success behind intelligent systems often relies on large-scale models with high computational complexity and storage costs. The over-parameterized networks are often easy to optimize and can achieve better performance. However, it is challenging to deploy them over resource-limited edge-devices. Knowledge Distillation (KD) aims to optimize a lightweight network from the perspective of over-parameterized training. The traditional offline KD transfers knowledge from a cumbersome teacher to a small and fast student network. When a sizeable pre-trained teacher network is unavailable, online KD can improve a group of models by collaborative or mutual learning. Without needing extra models, Self-KD boosts the network itself using attached auxiliary architectures. KD mainly involves knowledge extraction and distillation strategies these two aspects. Beyond KD schemes, various KD algorithms are widely used in practical applications, such as multi-teacher KD, cross-modal KD, attention-based KD, data-free KD and adversarial KD. This paper provides a comprehensive KD survey, including knowledge categories, distillation schemes and algorithms, as well as some empirical studies on performance comparison. Finally, we discuss the open challenges of existing KD works and prospect the future directions.