Pose distillation is widely adopted to reduce model size in human pose estimation. However, existing methods primarily emphasize the transfer of teacher knowledge while often neglecting the performance degradation resulted from the curse of capacity gap between teacher and student. To address this issue, we propose AgentPose, a novel pose distillation method that integrates a feature agent to model the distribution of teacher features and progressively aligns the distribution of student features with that of the teacher feature, effectively overcoming the capacity gap and enhancing the ability of knowledge transfer. Our comprehensive experiments conducted on the COCO dataset substantiate the effectiveness of our method in knowledge transfer, particularly in scenarios with a high capacity gap.
Transfer learning of prediction models has been extensively studied, while the corresponding policy learning approaches are rarely discussed. In this paper, we propose principled approaches for learning the optimal policy in the target domain by leveraging two datasets: one with full information from the source domain and the other from the target domain with only covariates. First, under the setting of covariate shift, we formulate the problem from a perspective of causality and present the identifiability assumptions for the reward induced by a given policy. Then, we derive the efficient influence function and the semiparametric efficiency bound for the reward. Based on this, we construct a doubly robust and semiparametric efficient estimator for the reward and then learn the optimal policy by optimizing the estimated reward. Moreover, we theoretically analyze the bias and the generalization error bound for the learned policy. Furthermore, in the presence of both covariate and concept shifts, we propose a novel sensitivity analysis method to evaluate the robustness of the proposed policy learning approach. Extensive experiments demonstrate that the approach not only estimates the reward more accurately but also yields a policy that closely approximates the theoretically optimal policy.
In this paper, we propose a novel reinforcement learning algorithm for inventory management of newly launched products with no historical demand information. The algorithm follows the classic Dyna-$Q$ structure, balancing the model-free and model-based approaches, while accelerating the training process of Dyna-$Q$ and mitigating the model discrepancy generated by the model-based feedback. Based on the idea of transfer learning, warm-start information from the demand data of existing similar products can be incorporated into the algorithm to further stabilize the early-stage training and reduce the variance of the estimated optimal policy. Our approach is validated through a case study of bakery inventory management with real data. The adjusted Dyna-$Q$ shows up to a 23.7\% reduction in average daily cost compared with $Q$-learning, and up to a 77.5\% reduction in training time within the same horizon compared with classic Dyna-$Q$. By using transfer learning, it can be found that the adjusted Dyna-$Q$ has the lowest total cost, lowest variance in total cost, and relatively low shortage percentages among all the benchmarking algorithms under a 30-day testing.
In nations such as Bangladesh, agriculture plays a vital role in providing livelihoods for a significant portion of the population. Identifying and classifying plant diseases early is critical to prevent their spread and minimize their impact on crop yield and quality. Various computer vision techniques can be used for such detection and classification. While CNNs have been dominant on such image classification tasks, vision transformers has become equally good in recent time also. In this paper we study the various computer vision techniques for Bangladeshi rice leaf disease detection. We use the Dhan-Shomadhan -- a Bangladeshi rice leaf disease dataset, to experiment with various CNN and ViT models. We also compared the performance of such deep neural network architecture with traditional machine learning architecture like Support Vector Machine(SVM). We leveraged transfer learning for better generalization with lower amount of training data. Among the models tested, ResNet50 exhibited the best performance over other CNN and transformer-based models making it the optimal choice for this task.
We define a digital twin (DT) of a physical system governed by partial differential equations (PDEs) as a model for real-time simulations and control of the system behavior under changing conditions. We construct DTs using the Karhunen-Lo\`{e}ve Neural Network (KL-NN) surrogate model and transfer learning (TL). The surrogate model allows fast inference and differentiability with respect to control parameters for control and optimization. TL is used to retrain the model for new conditions with minimal additional data. We employ the moment equations to analyze TL and identify parameters that can be transferred to new conditions. The proposed analysis also guides the control variable selection in DT to facilitate efficient TL. For linear PDE problems, the non-transferable parameters in the KL-NN surrogate model can be exactly estimated from a single solution of the PDE corresponding to the mean values of the control variables under new target conditions. Retraining an ML model with a single solution sample is known as one-shot learning, and our analysis shows that the one-shot TL is exact for linear PDEs. For nonlinear PDE problems, transferring of any parameters introduces errors. For a nonlinear diffusion PDE model, we find that for a relatively small range of control variables, some surrogate model parameters can be transferred without introducing a significant error, some can be approximately estimated from the mean-field equation, and the rest can be found using a linear residual least square problem or an ordinary linear least square problem if a small labeled dataset for new conditions is available. The former approach results in a one-shot TL while the latter approach is an example of a few-shot TL. Both methods are approximate for the nonlinear PDEs.
Recent feature masking knowledge distillation methods make use of attention mechanisms to identify either important spatial regions or channel clues for discriminative feature reconstruction. However, most of existing strategies perform global attention-guided feature masking distillation without delving into fine-grained visual clues in feature maps. In particular, uncovering locality-aware clues across different scales are conducive to reconstructing region-aware features, thereby significantly benefiting distillation performance. In this study, we propose a fine-grained adaptive feature masking distillation framework for accurate object detection. Different from previous methods in which global masking is performed on single-scale feature maps, we explore the scale-aware feature masking by performing feature distillation across various scales, such that the object-aware locality is encoded for improved feature reconstruction. In addition, our fine-grained feature distillation strategy is combined with a masking logits distillation scheme in which logits difference between teacher and student networks is utilized to guide the distillation process. Thus, it can help the student model to better learn from the teacher counterpart with improved knowledge transfer. Extensive experiments for detection task demonstrate the superiority of our method. For example, when RetinaNet, RepPoints and Cascade Mask RCNN are used as teacher detectors, the student network achieves mAP scores of 41.5\%, 42.9\%, and 42.6\%, respectively, outperforming state-of-the-art methods such as DMKD and FreeKD.
Predicting the impact of single-point amino acid mutations on protein stability is essential for understanding disease mechanisms and advancing drug development. Protein stability, quantified by changes in Gibbs free energy ($\Delta\Delta G$), is influenced by these mutations. However, the scarcity of data and the complexity of model interpretation pose challenges in accurately predicting stability changes. This study proposes the application of deep neural networks, leveraging transfer learning and fusing complementary information from different models, to create a feature-rich representation of the protein stability landscape. We developed four models, with our third model, ThermoMPNN+, demonstrating the best performance in predicting $\Delta\Delta G$ values. This approach, which integrates diverse feature sets and embeddings through latent transfusion techniques, aims to refine $\Delta\Delta G$ predictions and contribute to a deeper understanding of protein dynamics, potentially leading to advancements in disease research and drug discovery.
Despite the artificial intelligence (AI) revolution, deep learning has yet to achieve much success with tabular data due to heterogeneous feature space and limited sample sizes without viable transfer learning. The new era of generative AI, powered by large language models (LLM), brings unprecedented learning opportunities to diverse data and domains. This paper investigates the effectiveness of an LLM application programming interface (API) and transfer learning of LLM in tabular data classification. LLM APIs respond to input text prompts with tokenized data and instructions, whereas transfer learning finetunes an LLM for a target classification task. This paper proposes an end-to-end finetuning of LLM to demonstrate cross-data transfer learning on ten benchmark data sets when large pre-trained tabular data models do not exist to facilitate transfer learning. The proposed LLM finetuning method outperforms state-of-the-art machine and deep learning methods on tabular data with less than ten features - a standard feature size for tabular data sets. The transfer learning approach uses a fraction of the computational cost of other deep learning or API-based solutions while ensuring competitive or superior classification performance.
Large-N nationally representative surveys, which have profoundly shaped American politics scholarship, represent related but distinct domains -a key condition for transfer learning applications. These surveys are related through their shared demographic, party identification, and ideological variables, yet differ in that individual surveys often lack specific policy preference questions that researchers require. Our study introduces a novel application of transfer learning (TL) to address these gaps, marking the first systematic use of TL paradigms in the context of survey data. Specifically, models pre-trained on the Cooperative Election Study (CES) dataset are fine-tuned for use in the American National Election Studies (ANES) dataset to predict policy questions based on demographic variables. Even with a naive architecture, our transfer learning approach achieves approximately 92 percentage accuracy in predicting missing variables across surveys, demonstrating the robust potential of this method. Beyond this specific application, our paper argues that transfer learning is a promising framework for maximizing the utility of existing survey data. We contend that artificial intelligence, particularly transfer learning, opens new frontiers in social science methodology by enabling systematic knowledge transfer between well-administered surveys that share common variables but differ in their outcomes of interest.
Accurate human posture classification in images and videos is crucial for automated applications across various fields, including work safety, physical rehabilitation, sports training, or daily assisted living. Recently, multimodal learning methods, such as Contrastive Language-Image Pretraining (CLIP), have advanced significantly in jointly understanding images and text. This study aims to assess the effectiveness of CLIP in classifying human postures, focusing on its application in yoga. Despite the initial limitations of the zero-shot approach, applying transfer learning on 15,301 images (real and synthetic) with 82 classes has shown promising results. The article describes the full procedure for fine-tuning, including the choice for image description syntax, models and hyperparameters adjustment. The fine-tuned CLIP model, tested on 3826 images, achieves an accuracy of over 85%, surpassing the current state-of-the-art of previous works on the same dataset by approximately 6%, its training time being 3.5 times lower than what is needed to fine-tune a YOLOv8-based model. For more application-oriented scenarios, with smaller datasets of six postures each, containing 1301 and 401 training images, the fine-tuned models attain an accuracy of 98.8% and 99.1%, respectively. Furthermore, our experiments indicate that training with as few as 20 images per pose can yield around 90% accuracy in a six-class dataset. This study demonstrates that this multimodal technique can be effectively used for yoga pose classification, and possibly for human posture classification, in general. Additionally, CLIP inference time (around 7 ms) supports that the model can be integrated into automated systems for posture evaluation, e.g., for developing a real-time personal yoga assistant for performance assessment.
Recent advances have enabled heterogeneous multi-robot teams to learn complex and effective coordination skills. However, existing neural architectures that support heterogeneous teaming tend to force a trade-off between expressivity and efficiency. Shared-parameter designs prioritize sample efficiency by enabling a single network to be shared across all or a pre-specified subset of robots (via input augmentations), but tend to limit behavioral diversity. In contrast, recent designs employ a separate policy for each robot, enabling greater diversity and expressivity at the cost of efficiency and generalization. Our key insight is that such tradeoffs can be avoided by viewing these design choices as ends of a broad spectrum. Inspired by recent work in transfer and meta learning, and building on prior work in multi-robot task allocation, we propose Capability-Aware Shared Hypernetworks (CASH), a soft weight sharing architecture that uses hypernetworks to efficiently learn a flexible shared policy that dynamically adapts to each robot post-training. By explicitly encoding the impact of robot capabilities (e.g., speed and payload) on collective behavior, CASH enables zero-shot generalization to unseen robots or team compositions. Our experiments involve multiple heterogeneous tasks, three learning paradigms (imitation learning, value-based, and policy-gradient RL), and SOTA multi-robot simulation (JaxMARL) and hardware (Robotarium) platforms. Across all conditions, we find that CASH generates appropriately-diverse behaviors and consistently outperforms baseline architectures in terms of performance and sample efficiency during both training and zero-shot generalization, all with 60%-80% fewer learnable parameters.
Attribute classification is crucial for identifying specific characteristics within image regions. Vision-Language Models (VLMs) have been effective in zero-shot tasks by leveraging their general knowledge from large-scale datasets. Recent studies demonstrate that transformer-based models with class-wise queries can effectively address zero-shot multi-label classification. However, poor utilization of the relationship between seen and unseen attributes makes the model lack generalizability. Additionally, attribute classification generally involves many attributes, making maintaining the model's scalability difficult. To address these issues, we propose Super-class guided transFormer (SugaFormer), a novel framework that leverages super-classes to enhance scalability and generalizability for zero-shot attribute classification. SugaFormer employs Super-class Query Initialization (SQI) to reduce the number of queries, utilizing common semantic information from super-classes, and incorporates Multi-context Decoding (MD) to handle diverse visual cues. To strengthen generalizability, we introduce two knowledge transfer strategies that utilize VLMs. During training, Super-class guided Consistency Regularization (SCR) aligns model's features with VLMs using super-class guided prompts, and during inference, Zero-shot Retrieval-based Score Enhancement (ZRSE) refines predictions for unseen attributes. Extensive experiments demonstrate that SugaFormer achieves state-of-the-art performance across three widely-used attribute classification benchmarks under zero-shot, and cross-dataset transfer settings. Our code is available at https://github.com/mlvlab/SugaFormer.
We propose an efficient knowledge transfer approach for model-based reinforcement learning, addressing the challenge of deploying large world models in resource-constrained environments. Our method distills a high-capacity multi-task agent (317M parameters) into a compact 1M parameter model, achieving state-of-the-art performance on the MT30 benchmark with a normalized score of 28.45, a substantial improvement over the original 1M parameter model's score of 18.93. This demonstrates the ability of our distillation technique to consolidate complex multi-task knowledge effectively. Additionally, we apply FP16 post-training quantization, reducing the model size by 50% while maintaining performance. Our work bridges the gap between the power of large models and practical deployment constraints, offering a scalable solution for efficient and accessible multi-task reinforcement learning in robotics and other resource-limited domains.
This paper presents a novel approach for the automatic generation of Cued Speech (ACSG), a visual communication system used by people with hearing impairment to better elicit the spoken language. We explore transfer learning strategies by leveraging a pre-trained audiovisual autoregressive text-to-speech model (AVTacotron2). This model is reprogrammed to infer Cued Speech (CS) hand and lip movements from text input. Experiments are conducted on two publicly available datasets, including one recorded specifically for this study. Performance is assessed using an automatic CS recognition system. With a decoding accuracy at the phonetic level reaching approximately 77%, the results demonstrate the effectiveness of our approach.
Cloud removal plays a crucial role in enhancing remote sensing image analysis, yet accurately reconstructing cloud-obscured regions remains a significant challenge. Recent advancements in generative models have made the generation of realistic images increasingly accessible, offering new opportunities for this task. Given the conceptual alignment between image generation and cloud removal tasks, generative models present a promising approach for addressing cloud removal in remote sensing. In this work, we propose a deep transfer learning approach built on a generative adversarial network (GAN) framework to explore the potential of the novel masked autoencoder (MAE) image reconstruction model in cloud removal. Due to the complexity of remote sensing imagery, we further propose using a patch-wise discriminator to determine whether each patch of the image is real or not. The proposed reconstructive transfer learning approach demonstrates significant improvements in cloud removal performance compared to other GAN-based methods. Additionally, whilst direct comparisons with some of the state-of-the-art cloud removal techniques are limited due to unclear details regarding their train/test data splits, the proposed model achieves competitive results based on available benchmarks.
Accurate load forecasting is crucial for predictive control in many energy domain applications, with significant economic and ecological implications. To address these implications, this study provides an extensive benchmark of state-of-the-art deep learning models for short-term load forecasting in energy communities. Namely, LSTM, xLSTM, and Transformers are compared with benchmarks such as KNNs, synthetic load models, and persistence forecasting models. This comparison considers different scales of aggregation (e.g., number of household loads) and varying training data availability (e.g., training data time spans). Further, the impact of transfer learning from synthetic (standard) load profiles and the deep learning model size (i.e., parameter count) is investigated in terms of forecasting error. Implementations are publicly available and other researchers are encouraged to benchmark models using this framework. Additionally, a comprehensive case study, comprising an energy community of 50 households and a battery storage demonstrates the beneficial financial implications of accurate predictions. Key findings of this research include: (1) Simple persistence benchmarks outperform deep learning models for short-term load forecasting when the available training data is limited to six months or less; (2) Pretraining with publicly available synthetic load profiles improves the normalized Mean Absolute Error (nMAE) by an average of 1.28%pt during the first nine months of training data; (3) Increased aggregation significantly enhances the performance of deep learning models relative to persistence benchmarks; (4) Improved load forecasting, with an nMAE reduction of 1.1%pt, translates to an economic benefit of approximately 600EUR per year in an energy community comprising 50 households.
In the medical field, accurate diagnosis of lung cancer is crucial for treatment. Traditional manual analysis methods have significant limitations in terms of accuracy and efficiency. To address this issue, this paper proposes a deep learning network framework based on the pre-trained MobileNetV2 model, initialized with weights from the ImageNet-1K dataset (version 2). The last layer of the model (the fully connected layer) is replaced with a new fully connected layer, and a softmax activation function is added to efficiently classify three types of lung cancer CT scan images. Experimental results show that the model achieves an accuracy of 99.6% on the test set, with significant improvements in feature extraction compared to traditional models.With the rapid development of artificial intelligence technologies, deep learning applications in medical image processing are bringing revolutionary changes to the healthcare industry. AI-based lung cancer detection systems can significantly improve diagnostic efficiency, reduce the workload of doctors, and occupy an important position in the global healthcare market. The potential of AI to improve diagnostic accuracy, reduce medical costs, and promote precision medicine will have a profound impact on the future development of the healthcare industry.
As opposed to human drivers, current autonomous driving systems still require vast amounts of labeled data to train. Recently, world models have been proposed to simultaneously enhance autonomous driving capabilities by improving the way these systems understand complex real-world environments and reduce their data demands via self-supervised pre-training. In this paper, we present AD-L-JEPA (aka Autonomous Driving with LiDAR data via a Joint Embedding Predictive Architecture), a novel self-supervised pre-training framework for autonomous driving with LiDAR data that, as opposed to existing methods, is neither generative nor contrastive. Our method learns spatial world models with a joint embedding predictive architecture. Instead of explicitly generating masked unknown regions, our self-supervised world models predict Bird's Eye View (BEV) embeddings to represent the diverse nature of autonomous driving scenes. Our approach furthermore eliminates the need to manually create positive and negative pairs, as is the case in contrastive learning. AD-L-JEPA leads to simpler implementation and enhanced learned representations. We qualitatively and quantitatively demonstrate high-quality of embeddings learned with AD-L-JEPA. We furthermore evaluate the accuracy and label efficiency of AD-L-JEPA on popular downstream tasks such as LiDAR 3D object detection and associated transfer learning. Our experimental evaluation demonstrates that AD-L-JEPA is a plausible approach for self-supervised pre-training in autonomous driving applications and is the best available approach outperforming SOTA, including most recently proposed Occupancy-MAE [1] and ALSO [2]. The source code of AD-L-JEPA is available at https://github.com/HaoranZhuExplorer/AD-L-JEPA-Release.
In dynamic decision-making scenarios across business and healthcare, leveraging sample trajectories from diverse populations can significantly enhance reinforcement learning (RL) performance for specific target populations, especially when sample sizes are limited. While existing transfer learning methods primarily focus on linear regression settings, they lack direct applicability to reinforcement learning algorithms. This paper pioneers the study of transfer learning for dynamic decision scenarios modeled by non-stationary finite-horizon Markov decision processes, utilizing neural networks as powerful function approximators and backward inductive learning. We demonstrate that naive sample pooling strategies, effective in regression settings, fail in Markov decision processes.To address this challenge, we introduce a novel ``re-weighted targeting procedure'' to construct ``transferable RL samples'' and propose ``transfer deep $Q^*$-learning'', enabling neural network approximation with theoretical guarantees. We assume that the reward functions are transferable and deal with both situations in which the transition densities are transferable or nontransferable. Our analytical techniques for transfer learning in neural network approximation and transition density transfers have broader implications, extending to supervised transfer learning with neural networks and domain shift scenarios. Empirical experiments on both synthetic and real datasets corroborate the advantages of our method, showcasing its potential for improving decision-making through strategically constructing transferable RL samples in non-stationary reinforcement learning contexts.
Data-driven soft sensors (DDSS) have become mainstream methods for predicting key performance indicators in process industries. However, DDSS development requires complex and costly customized designs tailored to various tasks during the modeling process. Moreover, DDSS are constrained to a single structured data modality, limiting their ability to incorporate additional contextual knowledge. Furthermore, DDSSs' limited representation learning leads to weak predictive performance with scarce data. To address these challenges, we propose a general framework named LLM-TKESS (large language model for text-based knowledge-embedded soft sensing), harnessing the powerful general problem-solving capabilities, cross-modal knowledge transfer abilities, and few-shot capabilities of LLM for enhanced soft sensing modeling. Specifically, an auxiliary variable series encoder (AVS Encoder) is proposed to unleash LLM's potential for capturing temporal relationships within series and spatial semantic relationships among auxiliary variables. Then, we propose a two-stage fine-tuning alignment strategy: in the first stage, employing parameter-efficient fine-tuning through autoregressive training adjusts LLM to rapidly accommodate process variable data, resulting in a soft sensing foundation model (SSFM). Subsequently, by training adapters, we adapt the SSFM to various downstream tasks without modifying its architecture. Then, we propose two text-based knowledge-embedded soft sensors, integrating new natural language modalities to overcome the limitations of pure structured data models. Furthermore, benefiting from LLM's pre-existing world knowledge, our model demonstrates outstanding predictive capabilities in small sample conditions. Using the thermal deformation of air preheater rotor as a case study, we validate through extensive experiments that LLM-TKESS exhibits outstanding performance.