Search Results (16)

We describe AI agents as stochastic dynamical systems and frame the problem of learning to reason as in transductive inference: Rather than approximating the distribution of past data as in classical induction, the objective is to capture its algorithmic structure so as to reduce the time needed to solve new tasks. In this view, information from past experience serves not only to reduce a model’s uncertainty, as in Shannon’s classical theory, but to reduce the computational effort required to find solutions to unforeseen tasks. Working in the verifiable setting, where a checker or reward function is available, we establish three main results. First, we show that the optimal speed-up for a new task is tightly related to the algorithmic information it shares with the training data, yielding a theoretical justification for the power-law scaling empirically observed in reasoning models. Second, while the compression view of learning, rooted in Occam’s Razor, favors simplicity, we show that transductive inference yields its greatest benefits precisely when the data-generating mechanism is most complex. Third, we identify a possible failure mode of naïve scaling: in the limit of unbounded model size and computing, models with access to a reward signal can behave as savants, brute-forcing solutions without acquiring transferable reasoning strategies. Accordingly, we argue that a critical quantity to optimize when scaling reasoning models is time, the role of which in learning has remained largely unexplored. Full article

Alpha helical transmembrane proteins are a specific type of membrane proteins that consist of helices spanning the entire cell membrane. They make up almost a third of all transmembrane (TM) proteins and play a significant role in various cellular activities. The structural prediction of these proteins is crucial in understanding how they behave inside the cell and thus in identifying their functions. Despite their importance, only a small portion of TM proteins have had their structure determined experimentally. Inter-helical residue contact is one of the most successful computational approaches for reducing the TM protein fold search space and generating an acceptable 3D structure. Most current TM protein residue contact predictors use features extracted only from protein sequences to predict residue contacts. However, these features alone deliver a low-accuracy contact map and, as a result, a poor 3D structure. Although there are models that explore leveraging features extracted from protein 3D structures in order to produce a better representative contact model, such an approach remains theoretical, assuming the structure features are available, whereas in reality they are only available in the training data, but not in the test data, whose structure is what needs to be predicted. This presents a brand new transfer learning paradigm: training examples contain two sets of features, but test examples contain only one set of the less informative features. In this work, we propose a novel approach that can train a model with training examples that contain both sequence features and atomic features and apply the model on the test data that contain only sequence features but not atomic features, while still improving contact prediction rather than using sequence features alone. Specifically, our method, AT-TSVM, employs Active Transfer for Transductive Support Vector Machines, which is augmented with transfer, active learning and conventional transductive learning to enhance contact prediction accuracy. Results from a benchmark dataset show that our method can boost contact prediction accuracy by an average of 5 to 6% over the inductive classifier and 2.5 to 4% over the transductive classifier. Full article

The demand for safe, high-quality, and minimally processed food has intensified interest in non-destructive analytical techniques capable of assessing freshness and safety in real time. Among these, near-infrared (NIR) spectroscopy and biosensors have emerged as leading technologies due to their rapid, reagent-free, and sample-preserving nature. NIR spectroscopy offers a holistic assessment of internal compositional changes, while biosensors provide specific and sensitive detection of biological and chemical contaminants. Recent advances in miniaturization, chemometrics, and deep learning have further enhanced their potential for inline and point-of-need applications across diverse food matrices, including meat, seafood, eggs, fruits, and vegetables. This review critically evaluates the operational principles, instrumentation, and current applications of NIR spectroscopy and biosensors in food freshness and safety monitoring. It also explores their integration, highlights practical challenges such as calibration transfer and regulatory hurdles, and outlines emerging innovations including hybrid sensing, Artificial Intelligence (AI) integration, and smart packaging. The scope of this review is to provide a comprehensive understanding of these technologies, and its objective is to inform future research and industrial deployment strategies that support sustainable, real-time food quality control. These techniques enable near real-time monitoring under laboratory and pilot-scale conditions, showing strong potential for industrial adaptation. The nature of these targets often determines the choice of transduction method. Full article

Chemiresistive gas sensors have gained significant attention and have been widely applied in various fields. However, the gap between experimental observations and fundamental sensing mechanisms hinders systematic optimization. Despite the critical role of modeling in explaining atomic-scale interactions and offering predictive insights beyond experiments, existing reviews on chemiresistive gas sensors remain predominantly experimental-centric, with a limited systematic exploration of the modeling approaches. Herein, we present a comprehensive overview of the modeling approaches for chemiresistive gas sensors, focusing on two critical processes: the reception and transduction stages. For the reception process, density functional theory (DFT), molecular dynamics (MD), ab initio molecular dynamics (AIMD), and Monte Carlo (MC) methods were analyzed. DFT quantifies atomic-scale charge transfer, and orbital hybridization, MD/AIMD captures dynamic adsorption kinetics, and MC simulates equilibrium/non-equilibrium behaviors based on statistical mechanics principles. For the transduction process, band-bending-based theoretical models and power-law models elucidate the resistance modulation mechanisms, although their generalizability remains limited. Notably, the finite element method (FEM) has emerged as a powerful tool for full-process modeling by integrating gas diffusion, adsorption, and electronic responses into a unified framework. Future directions highlight the use of multiscale models to bridge microscopic interactions with macroscopic behaviors and the integration of machine learning, accelerating the iterative design of next-generation sensors with superior performance. Full article

Graph neural networks (GNNs) are crucial tools for processing non-Euclidean data. However, due to scalability issues caused by the dependency and topology of graph data, deploying GNNs in practical applications is challenging. Some methods aim to address this issue by transferring GNN knowledge to MLPs through knowledge distillation. However, distilled MLPs cannot directly capture graph structure information and rely only on node features, resulting in poor performance and sensitivity to noise. To solve this problem, we propose a lightweight optimization method for GNNs that combines graph contrastive learning and variable-temperature knowledge distillation. First, we use graph contrastive learning to capture graph structural representations, enriching the input information for the MLP. Then, we transfer GNN knowledge to the MLP using variable temperature knowledge distillation. Additionally, we enhance both node content and structural features before inputting them into the MLP, thus improving its performance and stability. Extensive experiments on seven datasets show that the proposed KDGCL model outperforms baseline models in both transductive and inductive settings; in particular, the KDGCL model achieves an average improvement of 1.63% in transductive settings and 0.8% in inductive settings when compared to baseline models. Furthermore, KDGCL maintains parameter efficiency and inference speed, making it competitive in terms of performance. Full article

Studying cell settlement in the three-dimensional structure of synthetic biomaterials over time is of great interest in research and clinical translation for the development of artificial tissues and organs. Tracking cells as physical objects improves our understanding of the processes of migration, homing, and cell division during colonisation of the artificial environment. In this study, the 3D environment had a direct effect on the behaviour of biological objects. Recently, deep learning-based algorithms have shown significant benefits for cell segmentation tasks and, furthermore, for biomaterial design optimisation. We analysed the primary LHON fibroblasts in an artificial 3D environment after adeno-associated virus transduction. Application of these tools to model cell homing in biomaterials and to monitor cell morphology, migration and proliferation indirectly demonstrated restoration of the normal cell phenotype after gene manipulation by AAV transduction. Following the 3Rs principles of reducing the use of living organisms in research, modeling the formation of tissues and organs by reconstructing the behaviour of different cell types on artificial materials facilitates drug testing, the study of inherited and inflammatory diseases, and wound healing. These studies on the composition and algorithms for creating biomaterials to model the formation of cell layers were inspired by the principles of biomimicry. Full article

Machine learning and digital health sensing data have led to numerous research achievements aimed at improving digital health technology. However, using machine learning in digital health poses challenges related to data availability, such as incomplete, unstructured, and fragmented data, as well as issues related to data privacy, security, and data format standardization. Furthermore, there is a risk of bias and discrimination in machine learning models. Thus, developing an accurate prediction model from scratch can be an expensive and complicated task that often requires extensive experiments and complex computations. Transfer learning methods have emerged as a feasible solution to address these issues by transferring knowledge from a previously trained task to develop high-performance prediction models for a new task. This survey paper provides a comprehensive study of the effectiveness of transfer learning for digital health applications to enhance the accuracy and efficiency of diagnoses and prognoses, as well as to improve healthcare services. The first part of this survey paper presents and discusses the most common digital health sensing technologies as valuable data resources for machine learning applications, including transfer learning. The second part discusses the meaning of transfer learning, clarifying the categories and types of knowledge transfer. It also explains transfer learning methods and strategies, and their role in addressing the challenges in developing accurate machine learning models, specifically on digital health sensing data. These methods include feature extraction, fine-tuning, domain adaptation, multitask learning, federated learning, and few-/single-/zero-shot learning. This survey paper highlights the key features of each transfer learning method and strategy, and discusses the limitations and challenges of using transfer learning for digital health applications. Overall, this paper is a comprehensive survey of transfer learning methods on digital health sensing data which aims to inspire researchers to gain knowledge of transfer learning approaches and their applications in digital health, enhance the current transfer learning approaches in digital health, develop new transfer learning strategies to overcome the current limitations, and apply them to a variety of digital health technologies. Full article

With the advent of deep learning and the accessibility of massive data, scene classification algorithms based on deep learning have been extensively researched and have achieved exciting developments. However, the success of deep models often relies on a large amount of annotated remote sensing data. Additionally, deep models are typically trained and tested on the same set of classes, leading to compromised generalization performance when encountering new classes. This is where few-shot learning aims to enable models to quickly generalize to new classes with only a few reference samples. In this paper, we propose a novel collaborative self-supervised transductive few-shot learning (CS${}^2$TFSL) algorithm for remote sensing scene classification. In our approach, we construct two distinct self-supervised auxiliary tasks to jointly train the feature extractor, aiming to obtain a powerful representation. Subsequently, the feature extractor’s parameters are frozen, requiring no further training, and transferred to the inference stage. During testing, we employ transductive inference to enhance the associative information between the support and query sets by leveraging additional sample information in the data. Extensive comparisons with state-of-the-art few-shot scene classification algorithms on the WHU-RS19 and NWPU-RESISC45 datasets demonstrate the effectiveness of the proposed CS${}^2$TFSL. More specifically, CS${}^2$TFSL ranks first in the settings of five-way one-shot and five-way five-shot. Additionally, detailed ablation experiments are conducted to analyze the CS${}^2$TFSL. The experimental results reveal significant and promising performance improvements in few-shot scene classification through the combination of self-supervised learning and direct transductive inference. Full article

In many real-life problems, it is difficult to acquire or label large amounts of data, resulting in so-called few-shot learning problems. However, few-shot classification is a challenging problem due to the uncertainty caused by using few labeled samples. In the past few years, many methods have been proposed with the common aim of transferring knowledge acquired on a previously solved task, which is often achieved by using a pretrained feature extractor. As such, if the initial task contains many labeled samples, it is possible to circumvent the limitations of few-shot learning. A shortcoming of existing methods is that they often require priors about the data distribution, such as the balance between considered classes. In this paper, we propose a novel transfer-based method with a double aim: providing state-of-the-art performance, as reported on standardized datasets in the field of few-shot learning, while not requiring such restrictive priors. Our methodology is able to cope with both inductive cases, where prediction is performed on test samples independently from each other, and transductive cases, where a joint (batch) prediction is performed. Full article

Precise and timely information on crop spatial distribution over large areas is paramount to agricultural monitoring, food security, and policy development. Currently, automatically classifying crop types at a large scale is challenging due to the scarcity of ground data. Although previous studies have indicated that transductive transfer learning (TTL) is a promising method to address this problem, it performs poorly within regions where crop compositions and phenology differ largely. Here we transferred random forest classifiers trained in limited regions with diversified growing conditions and land covers to the rest of the study area where ground data are scarce, with more than 130,000 Sentinel-2 images processed using the Google Earth Engine (GEE) platform. We established the 10 m crop maps for four major crops (i.e., maize, rapeseed, winter, and spring Triticeae crops) across 10 European Union (EU) countries from 2018 to 2019. The final crop maps had a high accuracy with overall accuracy generally greater than 0.89, with user’s accuracy and producer’s accuracy ranging from 0.72 to 0.98. Moreover, the resulting maps were consistent with the NUTS-2 level official statistics, with R2 consistently greater than 0.9. We further analyzed the crop rotation patterns and found that the rotation intervals across these EU countries were generally at least one year. Maize was dominantly rotated with winter Triticeae crops or converted to other land covers in the following year. Rapeseed was generally grown in rotation with winter Triticeae crops, whereas the rotation patterns of winter and spring Triticeae crops were more diversified. Red Edge Position (REP) and Normalized Difference Yellow Index (NDYI) played significant roles in crop classification across the EU. This study highlights the potential of the developed TTL method for crop classification over large spatial extents where labeled data are limited and the differences in crop compositions and phenology are relatively large. Full article

The prevalence of photosynthesis, as the major natural solar energy transduction mechanism or biophotovoltaics (BPV), has always intrigued mankind. Over the last decades, we have learned to extract this renewable energy through continuously improving solid-state semiconductive devices, such as the photovoltaic solar cell. Direct utilization of plant-based BPVs has, however, been almost impracticable so far. Nevertheless, the electrochemical platform of fuel cells (FCs) relying on redox potentials of algae suspensions or biofilms on functionalized anode materials has in recent years increasingly been demonstrated to produce clean or carbon-negative electrical power generators. Interestingly, these algal BPVs offer unparalleled advantages, including carbon sequestration, bioremediation and biomass harvesting, while producing electricity. The development of high performance and durable BPVs is dependent on upgraded anode materials with electrochemically dynamic nanostructures. However, the current challenges in the optimization of anode materials remain significant barriers towards the development of commercially viable technology. In this context, two-dimensional (2D) graphene-based carbonaceous material has widely been exploited in such FCs due to its flexible surface functionalization properties. Attempts to economically improve power outputs have, however, been futile owing to molecular scale disorders that limit efficient charge coupling for maximum power generation within the anodic films. Recently, Langmuir–Blodgett (LB) film has been substantiated as an efficacious film-forming technique to tackle the above limitations of algal BPVs; however, the aforesaid technology remains vastly untapped in BPVs. An in-depth electromechanistic view of the fabrication of LB films and their electron transference mechanisms is of huge significance for the scalability of BPVs. However, an inclusive review of LB films applicable to BPVs has yet to be undertaken, prohibiting futuristic applications. Consequently, we report an inclusive description of a contextual outline, functional principles, the LB film-formation mechanism, recent endeavors in developing LB films and acute encounters with prevailing BPV anode materials. Furthermore, the research and scale-up challenges relating to LB film-integrated BPVs are presented along with innovative perceptions of how to improve their practicability in scale-up processes. Full article

Signal drift caused by sensors or environmental changes, which can be regarded as data distribution changes over time, is related to transductive transfer learning, and the data in the target domain is not labeled. We propose a method that learns a subspace with maximum independence of the concentration features (MICF) according to the Hilbert-Schmidt Independence Criterion (HSIC), which reduces the inter-concentration discrepancy of distributions. Then, we use Iterative Fisher Linear Discriminant (IFLD) to extract the signal features by reducing the divergence within classes and increasing the divergence among classes, which helps to prevent inconsistent ratios of different types of samples among the domains. The effectiveness of MICF and IFLD was verified by three sets of experiments using sensors in real world conditions, along with experiments conducted in the authors’ laboratory. The proposed method achieved an accuracy of 76.17%, which was better than any of the existing methods that publish their data on a publicly available dataset (the Gas Sensor Drift Dataset). It was found that the MICF-IFLD was simple and effective, reduced interferences, and deftly managed tasks of transfer classification. Full article

The combination of transfer learning and remote sensing image processing technology can effectively improve the automation level of image information extraction from a remote sensing time series. However, in the processing of polarimetric synthetic aperture radar (PolSAR) time-series images, the existing transfer learning methods often cannot make full use of the time-series information of the images, relying too much on the labeled samples in the target domain. Furthermore, the speckle noise inherent in synthetic aperture radar (SAR) imagery aggravates the difficulty of the manual selection of labeled samples, so these methods have difficulty in meeting the processing requirements of large data volumes and high efficiency. In lieu of these problems and the spatio-temporal relational knowledge of objects in time-series images, this paper introduces the theory of time-series clustering and proposes a new three-phase time-series clustering algorithm. Due to the full use of the inherent characteristics of the PolSAR images, this algorithm can accurately transfer the labels of the source domain samples to those samples that have not changed in the whole time series without relying on the target domain labeled samples, so as to realize transductive sample label transfer for PolSAR time-series images. Experiments were carried out using three different sets of PolSAR time-series images and the proposed method was compared with two of the existing methods. The experimental results showed that the transfer precision of the proposed method reaches a high level with different data and different objects and it performs significantly better than the existing methods. With strong reliability and practicability, the proposed method can provide a new solution for the rapid information extraction of remote sensing image time series. Full article

Entropy measures have been a major interest of researchers to measure the information content of a dynamical system. One of the well-known methodologies is sample entropy, which is a model-free approach and can be deployed to measure the information transfer in time series. Sample entropy is based on the conditional entropy where a major concern is the number of past delays in the conditional term. In this study, we deploy a lag-specific conditional entropy to identify the informative past values. Moreover, considering the seasonality structure of data, we propose a clustering-based sample entropy to exploit the temporal information. Clustering-based sample entropy is based on the sample entropy definition while considering the clustering information of the training data and the membership of the test point to the clusters. In this study, we utilize the proposed method for transductive feature selection in black-box weather forecasting and conduct the experiments on minimum and maximum temperature prediction in Brussels for 1–6 days ahead. The results reveal that considering the local structure of the data can improve the feature selection performance. In addition, despite the large reduction in the number of features, the performance is competitive with the case of using all features. Full article