The landscape of video recognition has evolved significantly, shifting from traditional Convolutional Neural Networks (CNNs) to Transformer-based architectures for improved accuracy. While 3D CNNs have been effective at capturing spatiotemporal dynamics, recent Transformer models leverage self-attention to model long-range spatial and temporal dependencies. Despite achieving state-of-the-art performance on major benchmarks, Transformers remain computationally expensive, particularly with dense video data. To address this, we propose a lightweight Video Focal Modulation Network, DVFL-Net, which distills spatiotemporal knowledge from a large pre-trained teacher into a compact nano student model, enabling efficient on-device deployment. DVFL-Net utilizes knowledge distillation and spatial-temporal feature modulation to significantly reduce computation while preserving high recognition performance. We employ forward Kullback-Leibler (KL) divergence alongside spatio-temporal focal modulation to effectively transfer both local and global context from the Video-FocalNet Base (teacher) to the proposed VFL-Net (student). We evaluate DVFL-Net on UCF50, UCF101, HMDB51, SSV2, and Kinetics-400, benchmarking it against recent state-of-the-art methods in Human Action Recognition (HAR). Additionally, we conduct a detailed ablation study analyzing the impact of forward KL divergence. The results confirm the superiority of DVFL-Net in achieving an optimal balance between performance and efficiency, demonstrating lower memory usage, reduced GFLOPs, and strong accuracy, making it a practical solution for real-time HAR applications.
We often attribute human characteristics to large language models (LLMs) and claim that they "know" certain things. LLMs have an internal probabilistic knowledge that represents information retained during training. How can we assess the veracity of this knowledge? We examine two common methods for probing the veracity of LLMs and discover several assumptions that are flawed. To address these flawed assumptions, we introduce sAwMIL (short for Sparse Aware Multiple-Instance Learning), a probing method that utilizes the internal activations of LLMs to separate statements into true, false, and neither. sAwMIL is based on multiple-instance learning and conformal prediction. We evaluate sAwMIL on 5 validity criteria across 16 open-source LLMs, including both default and chat-based variants, as well as on 3 new datasets. Among the insights we provide are: (1) the veracity signal is often concentrated in the third quarter of an LLM's depth; (2) truth and falsehood signals are not always symmetric; (3) linear probes perform better on chat models than on default models; (4) nonlinear probes may be required to capture veracity signals for some LLMs with reinforcement learning from human feedback or knowledge distillation; and (5) LLMs capture a third type of signal that is distinct from true and false and is neither true nor false. These findings provide a reliable method for verifying what LLMs "know" and how certain they are of their probabilistic internal knowledge.
Visual SLAM is particularly challenging in environments affected by noise, varying lighting conditions, and darkness. Learning-based optical flow algorithms can leverage multiple modalities to address these challenges, but traditional optical flow-based visual SLAM approaches often require significant computational resources.To overcome this limitation, we propose FMF-SLAM, an efficient multimodal fusion SLAM method that utilizes fast Fourier transform (FFT) to enhance the algorithm efficiency. Specifically, we introduce a novel Fourier-based self-attention and cross-attention mechanism to extract features from RGB and depth signals. We further enhance the interaction of multimodal features by incorporating multi-scale knowledge distillation across modalities. We also demonstrate the practical feasibility of FMF-SLAM in real-world scenarios with real time performance by integrating it with a security robot by fusing with a global positioning module GNSS-RTK and global Bundle Adjustment. Our approach is validated using video sequences from TUM, TartanAir, and our real-world datasets, showcasing state-of-the-art performance under noisy, varying lighting, and dark conditions.Our code and datasets are available at https://github.com/youjie-zhou/FMF-SLAM.git.
Multimodal learning aims to leverage information from diverse data modalities to achieve more comprehensive performance. However, conventional multimodal models often suffer from modality imbalance, where one or a few modalities dominate model optimization, leading to suboptimal feature representation and underutilization of weak modalities. To address this challenge, we introduce Gradient-Guided Distillation (G$^{2}$D), a knowledge distillation framework that optimizes the multimodal model with a custom-built loss function that fuses both unimodal and multimodal objectives. G$^{2}$D further incorporates a dynamic sequential modality prioritization (SMP) technique in the learning process to ensure each modality leads the learning process, avoiding the pitfall of stronger modalities overshadowing weaker ones. We validate G$^{2}$D on multiple real-world datasets and show that G$^{2}$D amplifies the significance of weak modalities while training and outperforms state-of-the-art methods in classification and regression tasks. Our code is available at https://github.com/rAIson-Lab/G2D.
Self-supervised learning (SSL) models such as WavLM have brought substantial improvements to speaker diarization by providing rich contextual representations. However, the high computational and memory costs of these models hinder their deployment in real-time and resource-constrained scenarios. In this work, we present a comprehensive study on compressing SSL-based diarization models through structured pruning guided by knowledge distillation. Building upon our previous work, we extend the analysis to include pruning objectives based on multiply-accumulate operations (MACs), investigate module-wise and progressive pruning strategies, and examine the impact of training data quantity. Experimental results show that our method reduces model size by up to 80% without degrading performance, achieving up to 4x faster inference on a single GPU. We further perform large-scale evaluations on a diverse compound dataset comprising eight public diarization corpora, where our best pruned model achieves state-of-the-art performance across most conditions. Additionally, we show strong generalization to the CHiME-6 dataset, attaining performance comparable to the third-place system in the CHiME-7 challenge without any domain adaptation. All models and code are publicly released to support reproducibility and future research.
Federated learning aims to train a global model in a distributed environment that is close to the performance of centralized training. However, issues such as client label skew, data quantity skew, and other heterogeneity problems severely degrade the model's performance. Most existing methods overlook the scenario where only a small portion of clients participate in training within a large-scale client setting, whereas our experiments show that this scenario presents a more challenging federated learning task. Therefore, we propose a Knowledge Distillation with teacher-student Inequitable Aggregation (KDIA) strategy tailored to address the federated learning setting mentioned above, which can effectively leverage knowledge from all clients. In KDIA, the student model is the average aggregation of the participating clients, while the teacher model is formed by a weighted aggregation of all clients based on three frequencies: participation intervals, participation counts, and data volume proportions. During local training, self-knowledge distillation is performed. Additionally, we utilize a generator trained on the server to generate approximately independent and identically distributed (IID) data features locally for auxiliary training. We conduct extensive experiments on the CIFAR-10/100/CINIC-10 datasets and various heterogeneous settings to evaluate KDIA. The results show that KDIA can achieve better accuracy with fewer rounds of training, and the improvement is more significant under severe heterogeneity.
Federated learning (FL) is a decentralized collaborative machine learning (ML) technique. It provides a solution to the issues of isolated data islands and data privacy leakage in industrial ML practices. One major challenge in FL is handling the non-identical and independent distributed (non-IID) data. Current solutions either focus on constructing an all-powerful global model, or customizing personalized local models. Few of them can provide both a well-generalized global model and well-performed local models at the same time. Additionally, many FL solutions to the non-IID problem are benefited from introducing public datasets. However, this will also increase the risk of data leakage. To tackle the problems, we propose a novel data-free distillation framework, Federated Bidirectional Knowledge Distillation (FedBKD). Specifically, we train Generative Adversarial Networks (GAN) for synthetic data. During the GAN training, local models serve as discriminators and their parameters are frozen. The synthetic data is then used for bidirectional distillation between global and local models to achieve knowledge interactions so that performances for both sides are improved. We conduct extensive experiments on 4 benchmarks under different non-IID settings. The results show that FedBKD achieves SOTA performances in every case.
Graph Neural Networks (GNNs) have shown remarkable performance in the healthcare domain. However, what remained challenging is quantifying the predictive uncertainty of GNNs, which is an important aspect of trustworthiness in clinical settings. While Bayesian and ensemble methods can be used to quantify uncertainty, they are computationally expensive. Additionally, the disagreement metric used by ensemble methods to compute uncertainty cannot capture the diversity of models in an ensemble network. In this paper, we propose a novel method, based on knowledge distillation, to quantify GNNs' uncertainty more efficiently and with higher precision. We apply self-distillation, where the same network serves as both the teacher and student models, thereby avoiding the need to train several networks independently. To ensure the impact of self-distillation, we develop an uncertainty metric that captures the diverse nature of the network by assigning different weights to each GNN classifier. We experimentally evaluate the precision, performance, and ability of our approach in distinguishing out-of-distribution data on two graph datasets: MIMIC-IV and Enzymes. The evaluation results demonstrate that the proposed method can effectively capture the predictive uncertainty of the model while having performance similar to that of the MC Dropout and ensemble methods. The code is publicly available at https://github.com/tailabTMU/UQ_GNN.
Since self-attention layers in Transformers are permutation invariant by design, positional encodings must be explicitly incorporated to enable spatial understanding. However, fixed-size lookup tables used in traditional learnable position embeddings (PEs) limit extrapolation capabilities beyond pre-trained sequence lengths. Expert-designed methods such as ALiBi and RoPE, mitigate this limitation but demand extensive modifications for adapting to new modalities, underscoring fundamental challenges in adaptability and scalability. In this work, we present SeqPE, a unified and fully learnable position encoding framework that represents each $n$-dimensional position index as a symbolic sequence and employs a lightweight sequential position encoder to learn their embeddings in an end-to-end manner. To regularize SeqPE's embedding space, we introduce two complementary objectives: a contrastive objective that aligns embedding distances with a predefined position-distance function, and a knowledge distillation loss that anchors out-of-distribution position embeddings to in-distribution teacher representations, further enhancing extrapolation performance. Experiments across language modeling, long-context question answering, and 2D image classification demonstrate that SeqPE not only surpasses strong baselines in perplexity, exact match (EM), and accuracy--particularly under context length extrapolation--but also enables seamless generalization to multi-dimensional inputs without requiring manual architectural redesign. We release our code, data, and checkpoints at https://github.com/ghrua/seqpe.
Motion Object Segmentation (MOS) is crucial for autonomous driving, as it enhances localization, path planning, map construction, scene flow estimation, and future state prediction. While existing methods achieve strong performance, balancing accuracy and real-time inference remains a challenge. To address this, we propose a logits-based knowledge distillation framework for MOS, aiming to improve accuracy while maintaining real-time efficiency. Specifically, we adopt a Bird's Eye View (BEV) projection-based model as the student and a non-projection model as the teacher. To handle the severe imbalance between moving and non-moving classes, we decouple them and apply tailored distillation strategies, allowing the teacher model to better learn key motion-related features. This approach significantly reduces false positives and false negatives. Additionally, we introduce dynamic upsampling, optimize the network architecture, and achieve a 7.69% reduction in parameter count, mitigating overfitting. Our method achieves a notable IoU of 78.8% on the hidden test set of the SemanticKITTI-MOS dataset and delivers competitive results on the Apollo dataset. The KDMOS implementation is available at https://github.com/SCNU-RISLAB/KDMOS.
Adversarial Training (AT) is widely recognized as an effective approach to enhance the adversarial robustness of Deep Neural Networks. As a variant of AT, Adversarial Robustness Distillation (ARD) has shown outstanding performance in enhancing the robustness of small models. However, both AT and ARD face robust fairness issue: these models tend to display strong adversarial robustness against some classes (easy classes) while demonstrating weak adversarial robustness against others (hard classes). This paper explores the underlying factors of this problem and points out the smoothness degree of soft labels for different classes significantly impacts the robust fairness from both empirical observation and theoretical analysis. Based on the above exploration, we propose Anti-Bias Soft Label Distillation (ABSLD) within the Knowledge Distillation framework to enhance the adversarial robust fairness. Specifically, ABSLD adaptively reduces the student's error risk gap between different classes, which is accomplished by adjusting the class-wise smoothness degree of teacher's soft labels during the training process, and the adjustment is managed by assigning varying temperatures to different classes. Additionally, as a label-based approach, ABSLD is highly adaptable and can be integrated with the sample-based methods. Extensive experiments demonstrate ABSLD outperforms state-of-the-art methods on the comprehensive performance of robustness and fairness.
Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for training large language models (LLMs) on complex reasoning tasks, such as mathematical problem solving. A prerequisite for the scalability of RLVR is a high-quality problem set with precise and verifiable answers. However, the scarcity of well-crafted human-labeled math problems and limited-verification answers in existing distillation-oriented synthetic datasets limit their effectiveness in RL. Additionally, most problem synthesis strategies indiscriminately expand the problem set without considering the model's capabilities, leading to low efficiency in generating useful questions. To mitigate this issue, we introduce a Self-aware Weakness-driven problem Synthesis framework (SwS) that systematically identifies model deficiencies and leverages them for problem augmentation. Specifically, we define weaknesses as questions that the model consistently fails to learn through its iterative sampling during RL training. We then extract the core concepts from these failure cases and synthesize new problems to strengthen the model's weak areas in subsequent augmented training, enabling it to focus on and gradually overcome its weaknesses. Without relying on external knowledge distillation, our framework enables robust generalization byempowering the model to self-identify and address its weaknesses in RL, yielding average performance gains of 10.0% and 7.7% on 7B and 32B models across eight mainstream reasoning benchmarks.
Speech Emotion Recognition (SER) is crucial for improving human-computer interaction. Despite strides in monolingual SER, extending them to build a multilingual system remains challenging. Our goal is to train a single model capable of multilingual SER by distilling knowledge from multiple teacher models. To address this, we introduce a novel language-aware multi-teacher knowledge distillation method to advance SER in English, Finnish, and French. It leverages Wav2Vec2.0 as the foundation of monolingual teacher models and then distills their knowledge into a single multilingual student model. The student model demonstrates state-of-the-art performance, with a weighted recall of 72.9 on the English dataset and an unweighted recall of 63.4 on the Finnish dataset, surpassing fine-tuning and knowledge distillation baselines. Our method excels in improving recall for sad and neutral emotions, although it still faces challenges in recognizing anger and happiness.
Knowledge Distillation (KD) compresses large language models (LLMs) by transferring the teacher model's capabilities to a smaller student model, reducing inference cost and memory usage while maintaining performance. However, existing KD methods for LLMs often fail to prevent significant shifts in the student model's distribution during training, leading to issues such as catastrophic forgetting, mode collapse, and training-inference mismatch. To address these challenges, we propose a novel, plug-in curriculum learning framework inspired by the strength training principle of "progressive overload" (POCL), which can be seamlessly integrated into existing white-box KD approaches with minimal computational overhead. The framework comprises two core components: (1) a difficulty measurer that ranks and partitions training samples from easy to hard, and (2) a training scheduler that incrementally introduces these subsets into the distillation process at fixed intervals while applying loss functions with progressively rising temperatures. By starting with the easiest samples and progressively increasing the difficulty, the approach enhances both the stability and efficiency of learning. Extensive experiments in instruction-following settings demonstrate that POCL consistently improves the performance of distilled student models across various white-box KD methods and model families. Our findings highlight the effectiveness of sorted training samples in KD for LLMs. More generally, our work demonstrates how to structure training data within the KD process to enhance the stability and performance of distilled LLMs.
Single-stream architectures using Vision Transformer (ViT) backbones show great potential for real-time UAV tracking recently. However, frequent occlusions from obstacles like buildings and trees expose a major drawback: these models often lack strategies to handle occlusions effectively. New methods are needed to enhance the occlusion resilience of single-stream ViT models in aerial tracking. In this work, we propose to learn Occlusion-Robust Representations (ORR) based on ViTs for UAV tracking by enforcing an invariance of the feature representation of a target with respect to random masking operations modeled by a spatial Cox process. Hopefully, this random masking approximately simulates target occlusions, thereby enabling us to learn ViTs that are robust to target occlusion for UAV tracking. This framework is termed ORTrack. Additionally, to facilitate real-time applications, we propose an Adaptive Feature-Based Knowledge Distillation (AFKD) method to create a more compact tracker, which adaptively mimics the behavior of the teacher model ORTrack according to the task's difficulty. This student model, dubbed ORTrack-D, retains much of ORTrack's performance while offering higher efficiency. Extensive experiments on multiple benchmarks validate the effectiveness of our method, demonstrating its state-of-the-art performance. Codes is available at https://github.com/wuyou3474/ORTrack.
Model merging has emerged as a promising solution to accommodate multiple large models within constrained memory budgets. We present StatsMerging, a novel lightweight learning-based model merging method guided by weight distribution statistics without requiring ground truth labels or test samples. StatsMerging offers three key advantages: (1) It uniquely leverages singular values from singular value decomposition (SVD) to capture task-specific weight distributions, serving as a proxy for task importance to guide task coefficient prediction; (2) It employs a lightweight learner StatsMergeLearner to model the weight distributions of task-specific pre-trained models, improving generalization and enhancing adaptation to unseen samples; (3) It introduces Task-Specific Teacher Distillation for merging vision models with heterogeneous architectures, a merging learning paradigm that avoids costly ground-truth labels by task-specific teacher distillation. Notably, we present two types of knowledge distillation, (a) distilling knowledge from task-specific models to StatsMergeLearner; and (b) distilling knowledge from models with heterogeneous architectures prior to merging. Extensive experiments across eight tasks demonstrate the effectiveness of StatsMerging. Our results show that StatsMerging outperforms state-of-the-art techniques in terms of overall accuracy, generalization to unseen tasks, and robustness to image quality variations.
Class-Incremental Learning (CIL) aims to learn new classes sequentially while retaining the knowledge of previously learned classes. Recently, pre-trained models (PTMs) combined with parameter-efficient fine-tuning (PEFT) have shown remarkable performance in rehearsal-free CIL without requiring exemplars from previous tasks. However, existing adapter-based methods, which incorporate lightweight learnable modules into PTMs for CIL, create new adapters for each new task, leading to both parameter redundancy and failure to leverage shared knowledge across tasks. In this work, we propose ContinuaL Low-Rank Adaptation (CL-LoRA), which introduces a novel dual-adapter architecture combining \textbf{task-shared adapters} to learn cross-task knowledge and \textbf{task-specific adapters} to capture unique features of each new task. Specifically, the shared adapters utilize random orthogonal matrices and leverage knowledge distillation with gradient reassignment to preserve essential shared knowledge. In addition, we introduce learnable block-wise weights for task-specific adapters, which mitigate inter-task interference while maintaining the model's plasticity. We demonstrate CL-LoRA consistently achieves promising performance under multiple benchmarks with reduced training and inference computation, establishing a more efficient and scalable paradigm for continual learning with pre-trained models.
Multi-modal RGB and Depth (RGBD) data are predominant in many domains such as robotics, autonomous driving and remote sensing. The combination of these multi-modal data enhances environmental perception by providing 3D spatial context, which is absent in standard RGB images. Although RGBD multi-modal data can be available to train computer vision models, accessing all sensor modalities during the inference stage may be infeasible due to sensor failures or resource constraints, leading to a mismatch between data modalities available during training and inference. Traditional Cross-Modal Knowledge Distillation (CMKD) frameworks, developed to address this task, are typically based on a teacher/student paradigm, where a multi-modal teacher distills knowledge into a single-modality student model. However, these approaches face challenges in teacher architecture choices and distillation process selection, thus limiting their adoption in real-world scenarios. To overcome these issues, we introduce CroDiNo-KD (Cross-Modal Disentanglement: a New Outlook on Knowledge Distillation), a novel cross-modal knowledge distillation framework for RGBD semantic segmentation. Our approach simultaneously learns single-modality RGB and Depth models by exploiting disentanglement representation, contrastive learning and decoupled data augmentation with the aim to structure the internal manifolds of neural network models through interaction and collaboration. We evaluated CroDiNo-KD on three RGBD datasets across diverse domains, considering recent CMKD frameworks as competitors. Our findings illustrate the quality of CroDiNo-KD, and they suggest reconsidering the conventional teacher/student paradigm to distill information from multi-modal data to single-modality neural networks.
Respiratory sound datasets are limited in size and quality, making high performance difficult to achieve. Ensemble models help but inevitably increase compute cost at inference time. Soft label training distills knowledge efficiently with extra cost only at training. In this study, we explore soft labels for respiratory sound classification as an architecture-agnostic approach to distill an ensemble of teacher models into a student model. We examine different variations of our approach and find that even a single teacher, identical to the student, considerably improves performance beyond its own capability, with optimal gains achieved using only a few teachers. We achieve the new state-of-the-art Score of 64.39 on ICHBI, surpassing the previous best by 0.85 and improving average Scores across architectures by more than 1.16. Our results highlight the effectiveness of knowledge distillation with soft labels for respiratory sound classification, regardless of size or architecture.
Efficient deployment of large audio-language models for speech translation remains challenging due to their significant computational requirements. In this paper, we address this challenge through our system submissions to the "Model Compression" track at the International Conference on Spoken Language Translation (IWSLT 2025). We experiment with a combination of approaches including iterative layer pruning based on layer importance evaluation, low-rank adaptation with 4-bit quantization (QLoRA), and knowledge distillation. In our experiments, we use Qwen2-Audio-7B-Instruct for speech translation into German and Chinese. Our pruned (student) models achieve up to a 50% reduction in both model parameters and storage footprint, while retaining 97-100% of the translation quality of the in-domain (teacher) models.