Finding controllers that perform well across multiple morphologies is an important milestone for large-scale robotics, in line with recent advances via foundation models in other areas of machine learning. However, the challenges of learning a single controller to control multiple morphologies make the `one robot one task' paradigm dominant in the field. To alleviate these challenges, we present a pipeline that: (1) leverages Quality Diversity algorithms like MAP-Elites to create a dataset of many single-task/single-morphology teacher controllers, then (2) distills those diverse controllers into a single multi-morphology controller that performs well across many different body plans by mimicking the sensory-action patterns of the teacher controllers via supervised learning. The distilled controller scales well with the number of teachers/morphologies and shows emergent properties. It generalizes to unseen morphologies in a zero-shot manner, providing robustness to morphological perturbations and instant damage recovery. Lastly, the distilled controller is also independent of the teacher controllers -- we can distill the teacher's knowledge into any controller model, making our approach synergistic with architectural improvements and existing training algorithms for teacher controllers.
In this paper, we present a simple yet effective contrastive knowledge distillation approach, which can be formulated as a sample-wise alignment problem with intra- and inter-sample constraints. Unlike traditional knowledge distillation methods that concentrate on maximizing feature similarities or preserving class-wise semantic correlations between teacher and student features, our method attempts to recover the "dark knowledge" by aligning sample-wise teacher and student logits. Specifically, our method first minimizes logit differences within the same sample by considering their numerical values, thus preserving intra-sample similarities. Next, we bridge semantic disparities by leveraging dissimilarities across different samples. Note that constraints on intra-sample similarities and inter-sample dissimilarities can be efficiently and effectively reformulated into a contrastive learning framework with newly designed positive and negative pairs. The positive pair consists of the teacher's and student's logits derived from an identical sample, while the negative pairs are formed by using logits from different samples. With this formulation, our method benefits from the simplicity and efficiency of contrastive learning through the optimization of InfoNCE, yielding a run-time complexity that is far less than $O(n^2)$, where $n$ represents the total number of training samples. Furthermore, our method can eliminate the need for hyperparameter tuning, particularly related to temperature parameters and large batch sizes. We conduct comprehensive experiments on three datasets including CIFAR-100, ImageNet-1K, and MS COCO. Experimental results clearly confirm the effectiveness of the proposed method on both image classification and object detection tasks. Our source codes will be publicly available at https://github.com/wencheng-zhu/CKD.
Continual learning, the ability to acquire knowledge from new data while retaining previously learned information, is a fundamental challenge in machine learning. Various approaches, including memory replay, knowledge distillation, model regularization, and dynamic network expansion, have been proposed to address this issue. Thus far, dynamic network expansion methods have achieved state-of-the-art performance at the cost of incurring significant computational overhead. This is due to the need for additional model buffers, which makes it less feasible in resource-constrained settings, particularly in the medical domain. To overcome this challenge, we propose Dynamic Model Merging, DynaMMo, a method that merges multiple networks at different stages of model training to achieve better computational efficiency. Specifically, we employ lightweight learnable modules for each task and combine them into a unified model to minimize computational overhead. DynaMMo achieves this without compromising performance, offering a cost-effective solution for continual learning in medical applications. We evaluate DynaMMo on three publicly available datasets, demonstrating its effectiveness compared to existing approaches. DynaMMo offers around 10-fold reduction in GFLOPS with a small drop of 2.76 in average accuracy when compared to state-of-the-art dynamic-based approaches. The code implementation of this work will be available upon the acceptance of this work at https://github.com/BioMedIA-MBZUAI/DynaMMo.
Image classification is a fundamental task in computer vision, and the quest to enhance DNN accuracy without inflating model size or latency remains a pressing concern. We make a couple of advances in this regard, leading to a novel EncodeNet design and training framework. The first advancement involves Converting Autoencoders, a novel approach that transforms images into an easy-to-classify image of its class. Our prior work that applied the Converting Autoencoder and a simple classifier in tandem achieved moderate accuracy over simple datasets, such as MNIST and FMNIST. However, on more complex datasets like CIFAR-10, the Converting Autoencoder has a large reconstruction loss, making it unsuitable for enhancing DNN accuracy. To address these limitations, we generalize the design of Converting Autoencoders by leveraging a larger class of DNNs, those with architectures comprising feature extraction layers followed by classification layers. We incorporate a generalized algorithmic design of the Converting Autoencoder and intraclass clustering to identify representative images, leading to optimized image feature learning. Next, we demonstrate the effectiveness of our EncodeNet design and training framework, improving the accuracy of well-trained baseline DNNs while maintaining the overall model size. EncodeNet's building blocks comprise the trained encoder from our generalized Converting Autoencoders transferring knowledge to a lightweight classifier network - also extracted from the baseline DNN. Our experimental results demonstrate that EncodeNet improves the accuracy of VGG16 from 92.64% to 94.05% on CIFAR-10 and RestNet20 from 74.56% to 76.04% on CIFAR-100. It outperforms state-of-the-art techniques that rely on knowledge distillation and attention mechanisms, delivering higher accuracy for models of comparable size.
In this study, we focus on heterogeneous knowledge transfer across entirely different model architectures, tasks, and modalities. Existing knowledge transfer methods (e.g., backbone sharing, knowledge distillation) often hinge on shared elements within model structures or task-specific features/labels, limiting transfers to complex model types or tasks. To overcome these challenges, we present MergeNet, which learns to bridge the gap of parameter spaces of heterogeneous models, facilitating the direct interaction, extraction, and application of knowledge within these parameter spaces. The core mechanism of MergeNet lies in the parameter adapter, which operates by querying the source model's low-rank parameters and adeptly learning to identify and map parameters into the target model. MergeNet is learned alongside both models, allowing our framework to dynamically transfer and adapt knowledge relevant to the current stage, including the training trajectory knowledge of the source model. Extensive experiments on heterogeneous knowledge transfer demonstrate significant improvements in challenging settings, where representative approaches may falter or prove less applicable.
Subgraph federated learning (subgraph-FL) is a new distributed paradigm that facilitates the collaborative training of graph neural networks (GNNs) by multi-client subgraphs. Unfortunately, a significant challenge of subgraph-FL arises from subgraph heterogeneity, which stems from node and topology variation, causing the impaired performance of the global GNN. Despite various studies, they have not yet thoroughly investigated the impact mechanism of subgraph heterogeneity. To this end, we decouple node and topology variation, revealing that they correspond to differences in label distribution and structure homophily. Remarkably, these variations lead to significant differences in the class-wise knowledge reliability of multiple local GNNs, misguiding the model aggregation with varying degrees. Building on this insight, we propose topology-aware data-free knowledge distillation technology (FedTAD), enhancing reliable knowledge transfer from the local model to the global model. Extensive experiments on six public datasets consistently demonstrate the superiority of FedTAD over state-of-the-art baselines.
We show that Claude 3 Opus, a large language model (LLM) released by Anthropic in March 2024, exhibits stronger machine translation competence than other LLMs. Though we find evidence of data contamination with Claude on FLORES-200, we curate new benchmarks that corroborate the effectiveness of Claude for low-resource machine translation into English. We find that Claude has remarkable \textit{resource efficiency} -- the degree to which the quality of the translation model depends on a language pair's resource level. Finally, we show that advancements in LLM translation can be compressed into traditional neural machine translation (NMT) models. Using Claude to generate synthetic data, we demonstrate that knowledge distillation advances the state-of-the-art in Yoruba-English translation, meeting or surpassing strong baselines like NLLB-54B and Google Translate.
Temperature plays a pivotal role in moderating label softness in the realm of knowledge distillation (KD). Traditional approaches often employ a static temperature throughout the KD process, which fails to address the nuanced complexities of samples with varying levels of difficulty and overlooks the distinct capabilities of different teacher-student pairings. This leads to a less-than-ideal transfer of knowledge. To improve the process of knowledge propagation, we proposed Dynamic Temperature Knowledge Distillation (DTKD) which introduces a dynamic, cooperative temperature control for both teacher and student models simultaneously within each training iterafion. In particular, we proposed "\textbf{sharpness}" as a metric to quantify the smoothness of a model's output distribution. By minimizing the sharpness difference between the teacher and the student, we can derive sample-specific temperatures for them respectively. Extensive experiments on CIFAR-100 and ImageNet-2012 demonstrate that DTKD performs comparably to leading KD techniques, with added robustness in Target Class KD and None-target Class KD scenarios.The code is available at https://github.com/JinYu1998/DTKD.
Over the past year, the field of Natural Language Generation (NLG) has experienced an exponential surge, largely due to the introduction of Large Language Models (LLMs). These models have exhibited the most effective performance in a range of domains within the Natural Language Processing and Generation domains. However, their application in domain-specific tasks, such as paraphrasing, presents significant challenges. The extensive number of parameters makes them difficult to operate on commercial hardware, and they require substantial time for inference, leading to high costs in a production setting. In this study, we tackle these obstacles by employing LLMs to develop three distinct models for the paraphrasing field, applying a method referred to as sequence-level knowledge distillation. These distilled models are capable of maintaining the quality of paraphrases generated by the LLM. They demonstrate faster inference times and the ability to generate diverse paraphrases of comparable quality. A notable characteristic of these models is their ability to exhibit syntactic diversity while also preserving lexical diversity, features previously uncommon due to existing data quality issues in datasets and not typically observed in neural-based approaches. Human evaluation of our models shows that there is only a 4% drop in performance compared to the LLM teacher model used in the distillation process, despite being 1000 times smaller. This research provides a significant contribution to the NLG field, offering a more efficient and cost-effective solution for paraphrasing tasks.
Vertical Federated Learning (VFL) is a category of Federated Learning in which models are trained collaboratively among parties with vertically partitioned data. Typically, in a VFL scenario, the labels of the samples are kept private from all the parties except for the aggregating server, that is the label owner. Nevertheless, recent works discovered that by exploiting gradient information returned by the server to bottom models, with the knowledge of only a small set of auxiliary labels on a very limited subset of training data points, an adversary can infer the private labels. These attacks are known as label inference attacks in VFL. In our work, we propose a novel framework called KDk, that combines Knowledge Distillation and k-anonymity to provide a defense mechanism against potential label inference attacks in a VFL scenario. Through an exhaustive experimental campaign we demonstrate that by applying our approach, the performance of the analyzed label inference attacks decreases consistently, even by more than 60%, maintaining the accuracy of the whole VFL almost unaltered.
Data-free knowledge distillation (DFKD) is a promising approach for addressing issues related to model compression, security privacy, and transmission restrictions. Although the existing methods exploiting DFKD have achieved inspiring achievements in coarse-grained classification, in practical applications involving fine-grained classification tasks that require more detailed distinctions between similar categories, sub-optimal results are obtained. To address this issue, we propose an approach called DFKD-FGVC that extends DFKD to fine-grained visual categorization~(FGVC) tasks. Our approach utilizes an adversarial distillation framework with attention generator, mixed high-order attention distillation, and semantic feature contrast learning. Specifically, we introduce a spatial-wise attention mechanism to the generator to synthesize fine-grained images with more details of discriminative parts. We also utilize the mixed high-order attention mechanism to capture complex interactions among parts and the subtle differences among discriminative features of the fine-grained categories, paying attention to both local features and semantic context relationships. Moreover, we leverage the teacher and student models of the distillation framework to contrast high-level semantic feature maps in the hyperspace, comparing variances of different categories. We evaluate our approach on three widely-used FGVC benchmarks (Aircraft, Cars196, and CUB200) and demonstrate its superior performance.
The intensive computational burden of Stable Diffusion (SD) for text-to-image generation poses a significant hurdle for its practical application. To tackle this challenge, recent research focuses on methods to reduce sampling steps, such as Latent Consistency Model (LCM), and on employing architectural optimizations, including pruning and knowledge distillation. Diverging from existing approaches, we uniquely start with a compact SD variant, BK-SDM. We observe that directly applying LCM to BK-SDM with commonly used crawled datasets yields unsatisfactory results. It leads us to develop two strategies: (1) leveraging high-quality image-text pairs from leading generative models and (2) designing an advanced distillation process tailored for LCM. Through our thorough exploration of quantization, profiling, and on-device deployment, we achieve rapid generation of photo-realistic, text-aligned images in just two steps, with latency under one second on resource-limited edge devices.
Pre-trained vision-language (V-L) models such as CLIP have shown excellent performance in many downstream cross-modal tasks. However, most of them are only applicable to the English context. Subsequent research has focused on this problem and proposed improved models, such as CN-CLIP and AltCLIP, to facilitate their applicability to Chinese and even other languages. Nevertheless, these models suffer from high latency and a large memory footprint in inference, which limits their further deployment on resource-constrained edge devices. In this work, we propose a conceptually simple yet effective multilingual CLIP Compression framework and train a lightweight multilingual vision-language model, called DC-CLIP, for both Chinese and English context. In this framework, we collect high-quality Chinese and English text-image pairs and design two training stages, including multilingual vision-language feature distillation and alignment. During the first stage, lightweight image/text student models are designed to learn robust visual/multilingual textual feature representation ability from corresponding teacher models, respectively. Subsequently, the multilingual vision-language alignment stage enables effective alignment of visual and multilingual textual features to further improve the model's multilingual performance. Comprehensive experiments in zero-shot image classification, conducted based on the ELEVATER benchmark, showcase that DC-CLIP achieves superior performance in the English context and competitive performance in the Chinese context, even with less training data, when compared to existing models of similar parameter magnitude. The evaluation demonstrates the effectiveness of our designed training mechanism.
Compact neural networks are specially designed for applications on edge devices with faster inference speed yet modest performance. However, training strategies of compact models are borrowed from that of conventional models at present, which ignores their difference in model capacity and thus may impede the performance of compact models. In this paper, by systematically investigating the impact of different training ingredients, we introduce a strong training strategy for compact models. We find that the appropriate designs of re-parameterization and knowledge distillation are crucial for training high-performance compact models, while some commonly used data augmentations for training conventional models, such as Mixup and CutMix, lead to worse performance. Our experiments on ImageNet-1K dataset demonstrate that our specialized training strategy for compact models is applicable to various architectures, including GhostNetV2, MobileNetV2 and ShuffleNetV2. Specifically, equipped with our strategy, GhostNetV3 1.3$\times$ achieves a top-1 accuracy of 79.1% with only 269M FLOPs and a latency of 14.46ms on mobile devices, surpassing its ordinarily trained counterpart by a large margin. Moreover, our observation can also be extended to object detection scenarios. PyTorch code and checkpoints can be found at https://github.com/huawei-noah/Efficient-AI-Backbones/tree/master/ghostnetv3_pytorch.
In the era of AIGC, the demand for low-budget or even on-device applications of diffusion models emerged. In terms of compressing the Stable Diffusion models (SDMs), several approaches have been proposed, and most of them leveraged the handcrafted layer removal methods to obtain smaller U-Nets, along with knowledge distillation to recover the network performance. However, such a handcrafting manner of layer removal is inefficient and lacks scalability and generalization, and the feature distillation employed in the retraining phase faces an imbalance issue that a few numerically significant feature loss terms dominate over others throughout the retraining process. To this end, we proposed the layer pruning and normalized distillation for compressing diffusion models (LAPTOP-Diff). We, 1) introduced the layer pruning method to compress SDM's U-Net automatically and proposed an effective one-shot pruning criterion whose one-shot performance is guaranteed by its good additivity property, surpassing other layer pruning and handcrafted layer removal methods, 2) proposed the normalized feature distillation for retraining, alleviated the imbalance issue. Using the proposed LAPTOP-Diff, we compressed the U-Nets of SDXL and SDM-v1.5 for the most advanced performance, achieving a minimal 4.0% decline in PickScore at a pruning ratio of 50% while the comparative methods' minimal PickScore decline is 8.2%. We will release our code.
We present a scalable framework designed to craft efficient lightweight models for video object detection utilizing self-training and knowledge distillation techniques. We scrutinize methodologies for the ideal selection of training images from video streams and the efficacy of model sharing across numerous cameras. By advocating for a camera clustering methodology, we aim to diminish the requisite number of models for training while augmenting the distillation dataset. The findings affirm that proper camera clustering notably amplifies the accuracy of distilled models, eclipsing the methodologies that employ distinct models for each camera or a universal model trained on the aggregate camera data.
Vision Transformers (ViT) have marked a paradigm shift in computer vision, outperforming state-of-the-art models across diverse tasks. However, their practical deployment is hampered by high computational and memory demands. This study addresses the challenge by evaluating four primary model compression techniques: quantization, low-rank approximation, knowledge distillation, and pruning. We methodically analyze and compare the efficacy of these techniques and their combinations in optimizing ViTs for resource-constrained environments. Our comprehensive experimental evaluation demonstrates that these methods facilitate a balanced compromise between model accuracy and computational efficiency, paving the way for wider application in edge computing devices.
In recent years, multimodal Graph Convolutional Networks (GCNs) have achieved remarkable performance in skeleton-based action recognition. The reliance on high-energy-consuming continuous floating-point operations inherent in GCN-based methods poses significant challenges for deployment in energy-constrained, battery-powered edge devices. To address these limitations, MK-SGN, a Spiking Graph Convolutional Network with Multimodal Fusion and Knowledge Distillation, is proposed to leverage the energy efficiency of Spiking Neural Networks (SNNs) for skeleton-based action recognition for the first time. By integrating the energy-saving properties of SNNs with the graph representation capabilities of GCNs, MK-SGN achieves significant reductions in energy consumption while maintaining competitive recognition accuracy. Firstly, we formulate a Spiking Multimodal Fusion (SMF) module to effectively fuse multimodal skeleton data represented as spike-form features. Secondly, we propose the Self-Attention Spiking Graph Convolution (SA-SGC) module and the Spiking Temporal Convolution (STC) module, to capture spatial relationships and temporal dynamics of spike-form features. Finally, we propose an integrated knowledge distillation strategy to transfer information from the multimodal GCN to the SGN, incorporating both intermediate-layer distillation and soft-label distillation to enhance the performance of the SGN. MK-SGN exhibits substantial advantages, surpassing state-of-the-art GCN frameworks in energy efficiency and outperforming state-of-the-art SNN frameworks in recognition accuracy. The proposed method achieves a remarkable reduction in energy consumption, exceeding 98\% compared to conventional GCN-based approaches. This research establishes a robust baseline for developing high-performance, energy-efficient SNN-based models for skeleton-based action recognition
In this research, we propose an innovative method to boost Knowledge Distillation efficiency without the need for resource-heavy teacher models. Knowledge Distillation trains a smaller ``student'' model with guidance from a larger ``teacher'' model, which is computationally costly. However, the main benefit comes from the soft labels provided by the teacher, helping the student grasp nuanced class similarities. In our work, we propose an efficient method for generating these soft labels, thereby eliminating the need for a large teacher model. We employ a compact autoencoder to extract essential features and calculate similarity scores between different classes. Afterward, we apply the softmax function to these similarity scores to obtain a soft probability vector. This vector serves as valuable guidance during the training of the student model. Our extensive experiments on various datasets, including CIFAR-100, Tiny Imagenet, and Fashion MNIST, demonstrate the superior resource efficiency of our approach compared to traditional knowledge distillation methods that rely on large teacher models. Importantly, our approach consistently achieves similar or even superior performance in terms of model accuracy. We also perform a comparative study with various techniques recently developed for knowledge distillation showing our approach achieves competitive performance with using significantly less resources. We also show that our approach can be easily added to any logit based knowledge distillation method. This research contributes to making knowledge distillation more accessible and cost-effective for practical applications, making it a promising avenue for improving the efficiency of model training. The code for this work is available at, https://github.com/JEKimLab/ReffAKD.
Recently, diffusion-based depth estimation methods have drawn widespread attention due to their elegant denoising patterns and promising performance. However, they are typically unreliable under adverse conditions prevalent in real-world scenarios, such as rainy, snowy, etc. In this paper, we propose a novel robust depth estimation method called D4RD, featuring a custom contrastive learning mode tailored for diffusion models to mitigate performance degradation in complex environments. Concretely, we integrate the strength of knowledge distillation into contrastive learning, building the `trinity' contrastive scheme. This scheme utilizes the sampled noise of the forward diffusion process as a natural reference, guiding the predicted noise in diverse scenes toward a more stable and precise optimum. Moreover, we extend noise-level trinity to encompass more generic feature and image levels, establishing a multi-level contrast to distribute the burden of robust perception across the overall network. Before addressing complex scenarios, we enhance the stability of the baseline diffusion model with three straightforward yet effective improvements, which facilitate convergence and remove depth outliers. Extensive experiments demonstrate that D4RD surpasses existing state-of-the-art solutions on synthetic corruption datasets and real-world weather conditions. Source code and data are available at \url{https://github.com/wangjiyuan9/D4RD}.