Space Representation Research Articles

Augmentation-aware self-supervised learning with conditioned projector

Self-supervised learning (SSL) is a powerful technique for learning from unlabeled data. By learning to remain invariant to applied data augmentations, methods such as SimCLR and MoCo can reach quality on par with supervised approaches. However, this invariance may be detrimental for solving downstream tasks that depend on traits affected by augmentations used during pretraining, such as color. In this paper, we propose to foster sensitivity to such characteristics in the representation space by modifying the projector network, a common component of self-supervised architectures. Specifically, we supplement the projector with information about augmentations applied to images. For the projector to take advantage of this auxiliary conditioning when solving the SSL task, the feature extractor learns to preserve the augmentation information in its representations. Our approach, coined Conditional Augmentation-aware Self-supervised Learning (CASSLE), is directly applicable to typical joint-embedding SSL methods regardless of their objective functions. Moreover, it does not require major changes in the network architecture or prior knowledge of downstream tasks. In addition to an analysis of sensitivity towards different data augmentations, we conduct a series of experiments, which show that CASSLE improves over various SSL methods, reaching state-of-the-art performance in multiple downstream tasks.11A short version of this paper appeared at the NeurIPS 2023 Workshop: Self-Supervised Learning - Theory and Practice (https://sslneurips23.github.io).

Multiple Heterogeneous Networks Representation With Latent Space for Synthetic Lethality Prediction.

Computational synthetic lethality (SL) method has become a promising strategy to identify SL gene pairs for targeted cancer therapy and cancer medicine development. Feature representation for integrating various biological networks is crutial to improve the identification performance. However, previous feature representation, such as matrix factorization and graph neural network, projects gene features onto latent variables by keeping a specific geometric metric. There is a lack of models of gene representational latent space with considerating multiple dimentionalities correlation and preserving latent geometric structures in both sample and feature spaces. Therefore, we propose a novel method to model gene Latent Space using matrix Tri-Factorization (LSTF) to obtain gene representation with embedding variables resulting from the potential interpretation of synthetic lethality. Meanwhile, manifold subspace regularization is applied to the tri-factorization to capture the geometrical manifold structure in the latent space with gene PPI functional and GO semantic embeddings. Then, SL gene pairs are identified by the reconstruction of the associations with gene representations in the latent space. The experimental results illustrate that LSTF is superior to other state-of-the-art methods. Case study demonstrate the effectiveness of the predicted SL associations.

Sifting the debris: Patterns in the SNR population with unsupervised ML methods

Context. Supernova remnants (SNRs) carry vast amounts of mechanical and radiative energy that heavily influence the structural, dynamical, and chemical evolution of galaxies. To this day, more than 300 SNRs have been discovered in the Milky Way, exhibiting a wide variety of observational features. However, existing classification schemes are mainly based on their radio morphology. Aims. In this work, we introduce a novel unsupervised deep learning pipeline to analyse a representative subsample of the Galactic SNR population (~50% of the total) with the aim of finding a connection between their multi-wavelength features and their physical properties. Methods. The pipeline involves two stages: (1) a representation learning stage, consisting of a convolutional autoencoder that feeds on imagery from infrared and radio continuum surveys (WISE 22 μm, Hi-GAL 70 μm and SMGPS 30 cm) and produces a compact representation in a lower-dimensionality latent space; and (2) a clustering stage that seeks meaningful clusters in the latent space that can be linked to the physical properties of the SNRs and their surroundings. Results. Our results suggest that this approach, when combined with an intermediate uniform manifold approximation and projection (UMAP) reprojection of the autoencoded embeddings into a more clusterable manifold, enables us to find reliable clusters. Despite a large number of sources being classified as outliers, most clusters relate to the presence of distinctive features, such as the distribution of infrared emission, the presence of radio shells and pulsar wind nebulae, and the existence of dust filaments.

Fine-grained detoxification framework via instance-level prefixes for large language models

Large Language Models (LLMs) have demonstrated remarkable performance across various natural language processing (NLP) tasks. However, their practical usability is often compromised by a propensity to generate toxic content, such as insults, threats, and profanity, particularly in response to adversarial prompts. Several fine-tuning and decoding approaches have been employed to address this challenge to mitigate toxicity. Nevertheless, these methods typically necessitate additional resources, such as high-quality training data or auxiliary models, thereby incurring extra costs. In this paper, we propose a novel detoxification framework, Fine-Grained Detoxification via Instance-Level Prefixes (FGDILP), which effectively mitigates the generation of toxic text without incurring additional training costs. Specifically, FGDILP leverages contextualized representations in attention space by contrasting a positive prefix-prepended prompt against multiple negative prefix-prepended prompts at the instance level. This methodology facilitates the construction of fine-grained subtoxicity vectors, which are subsequently fused to adjust the original generation pathway when responding to raw prompts. Our results demonstrate that FGDILP enables controlled text generation concerning detoxification at both the utterance and context levels. While our method slightly impacts generation fluency and diversity, it significantly outperforms prompt-based baselines regarding detoxification effectiveness. Our code is available at https://github.com/xinykou/FGDILP.

Shadow Representation: Making Claims to Represent Better Than the Official Representative

ABSTRACT This article advances knowledge on the diverse ways contemporary political representation is being performed. It identifies a particular kind of claim-making, shadow representation, whereby an individual or group makes claims to represent better than the officially authorised spokesperson or group. Shadow representation is enacted outside or alongside authorised spaces of representation by diverse actors such as civil society groups, lobbyists, political challengers, and former incumbents. The article conceptualises shadow representation in broad terms, and then examines its enactment in the electoral realm. The empirical analysis centres on two Australian cases in which defeated electoral candidates continue to make representative claims post-election. In both instances the shadow representatives construct a neglected constituency, and make claims to rectify representative deficiencies in the elected official. Their claim-making occurs between elections, and draws legitimacy from their strong community and voter support – two features that set them apart from the traditional political challenger focused on ousting an incumbent in an election. The article reflects on the democratic implications of shadow representation in electoral democracy: it potentially boosts the public scrutiny and accountability of elected officials, but it can also be used to destabilise the legitimacy of electoral procedures and their outcomes.

Heterogeneously Networked Evolutionary Games With Intergroup Conflicts.

Network games primarily explore the intricacies of individual interactions and adaptive strategies within a network. Building upon this framework, the present study delves into the modeling, analysis, and control of heterogeneously networked evolutionary games with intergroup conflict (HNEG-IC), where attacking players possess area-monitoring capabilities with limited attacking power. To begin with, a mathematical model is introduced to capture intragroup strategy dynamics and intergroup conflicts of HNEGs-IC via the algebraic state space representation (ASSR). A necessary and sufficient condition for achieving global cooperation of HNEGs-IC is established. Then, a criterion for verifying the κ -cooperation below a certain mortality is presented. Considering the HNEGs-IC with strategy feedback control, it is proven that the feedback control, subject to global cooperation, is robust to conflicts when the intersection of the strategy threshold set and the reachable set of the preset initial strategy profiles is empty. Finally, for verification and demonstration, the obtained results are applied to a simplified virtual game model of the NATO and the Warsaw Pact.

Non-degenerate metrics, hypersurface deformation algebra, non-anomalous representations and density weights in quantum gravity

Classical General Relativity is a dynamical theory of spacetime metrics of Lorentzian signature. In particular the classical metric field is nowhere degenerate in spacetime. In its initial value formulation with respect to a Cauchy surface the induced metric is of Euclidian signature and nowhere degenerate on it. It is only under this assumption of non-degeneracy of the induced metric that one can derive the hypersurface deformation algebra between the initial value constraints which is absolutely transparent from the fact that the inverse of the induced metric is needed to close the algebra. This statement is independent of the density weight that one may want to equip the spatial metric with. Accordingly, the very definition of a non-anomalous representation of the hypersurface deformation algebra in quantum gravity has to address the issue of non-degeneracy of the induced metric that is needed in the classical theory. In the Hilbert space representation employed in Loop Quantum Gravity (LQG) most emphasis has been laid to define an inverse metric operator on the dense domain of spin network states although they represent induced quantum geometries which are degenerate almost everywhere. It is no surprise that demonstration of closure of the constraint algebra on this domain meets difficulties because it is a sector of the quantum theory which is classically forbidden and which lies outside the domain of definition of the classical hypersurface deformation algebra. Various suggestions for addressing the issue such as non-standard operator topologies, dual spaces (habitats) and density weights have been proposed to address this issue with respect to the quantum dynamics of LQG. In this article we summarise these developments and argue that insisting on a dense domain of non-degenerate states within the LQG representation may provide a natural resolution of the issue thereby possibly avoiding the above mentioned non-standard constructions.

Data-Driven Modeling of DC–DC Power Converters

This article presents a data-driven methodology for modeling DC–DC power electronic converters. Using the proposed methodology, the dynamics of a converter can be captured, thereby eliminating the need for explicit theoretical modeling methods. This approach only requires the acquisition of fundamental measurements: currents through inductors and voltages across capacitors. The acquired data are used to construct a linear difference system, which is algebraically manipulated to form a state–space representation of the converter under analysis. Three DC–DC converter topologies were analyzed, and their resulting models were tested and compared with simulation data, yielding an average error deviation of approximately 2% for current signals and 4% for voltage signals, demonstrating precise tracking of the actual dynamics. The proposed data-driven methodology could simplify the implementation of adaptive control strategies in larger-scale solutions or in the interconnection of multiple converters.

Encoder decoder-based Virtual Physically Unclonable Function for Internet of Things device authentication using split-learning

Internet of Things (IoT) networks have been deployed widely making device authentication a crucial requirement that poses challenges related to security vulnerabilities, power consumption, and maintenance overheads. While current cryptographic techniques secure device communication; storing keys in Non-Volatile Memory (NVM) poses challenges for edge devices. Physically Unclonable Functions (PUFs) offer robust hardware-based authentication but introduce complexities such as hardware production and conservation expenses and susceptibility to aging effects. This paper’s main contribution is a novel scheme based on split learning, utilizing an encoder–decoder architecture at the device and server nodes to first create a Virtual PUF (VPUF) that addresses the shortcomings of the hardware PUF and secondly perform device authentication. The proposed VPUF reduces maintenance and power demands compared to the hardware PUF while enhancing security by transmitting latent space representations of responses between the node and the server. Also, since the encoder is placed on the node, while the decoder is on the server, this approach further reduces the computational load and processing time on the resource-constrained node. The obtained results demonstrate the effectiveness of the proposed VPUF scheme in modeling the behavior of the hardware-based PUF. Additionally, we investigate the impact of Gaussian noise in the communication channel between the server and the node on the system performance. The obtained results further reveal that the achieved authentication accuracy of the proposed scheme is 100%, as measured by the validation rate of the legitimate nodes. This highlights the superior performance of the proposed scheme in emulating the capabilities of a hardware-based PUF while providing secure and efficient authentication in IoT networks.

Dynamical phase transitions, caustics, and quantum dark bands

Abstract We provide a new perspective on quantum dynamical phase transitions (DPTs) by explaining their origin in terms of caustics that form in the Fock space representation of the many-body state over time, using the fully connected transverse field Ising model as an example. In this way we establish a connection between DPTs in a quantum spin system and an everyday natural phenomenon: The dark band between the primary and secondary bows (caustics) in rainbows known as Alexander’s dark band. The DPT occurs when the Loschmidt echo crosses the switching line between the evanescent tails of two back-to-back Airy functions that dress neighbouring fold caustics in Fock space and is the time-dependent analogue of what is seen as a function of angle in the sky. The structural stability and universal properties of caustics, as described mathematically by catastrophe theory, explains the generic occurrence of DPTs in the model and suggests that our analysis has wide applicability. Based on our thorough analytical understanding we propose a protocol which can be used to verify the existence of a DPT in a finite system experiment.

The Representations of Space and Transcendental Recognition in the Magazine Baikjo

The Representations of Space and Transcendental Recognition in the Magazine Baikjo

Textile Collage as the Artistic Method to Comprehend Industrial Memory: the Analysis of the Project “Radiant City” by Elena Sharganova

The article is focused on the sociological analysis of the project “Radiant city” of Elena Sharganova, resident of the tenth season of Open Studios at the Center of Contemporary Art Winzavod (2023/24), that is devoted to the artist’s hometown — Nelidovo — and consists of a series of textile collages, made with mixed media techniques including photo printing, cyanotype, machine seam, hand embroidery. Sociological conceptualization of the project is made based on the interview with the artist, analysis of archive publications kindly provided by the artist, and researches on social memory and industrial cities. Based on the interpretation of collages as spaces of representation in terms of Henri Lefebvre that represent a zone of experience, in which images and symbols fill in physical space and knowledge about it, the article shows that the “Radiant city” is the artist’s reflections on Nelidovo’s unrealized development trajectories and destinies of small industrial cities in general. There are three key topics raised by the artist in this project: disappointment in communism and nostalgia for the past, the problem of small industrial cities’ development, interrelation of collective and individual memory. Elena Sharganova’s project significance is defined by experiments with art mediums and a research approach that brings together art and social issues.

THE ESSENCE AND SCOPE OF UKRAINE'S VICTORY IN THE CONTEXT OF RUSSIA’S TOTAL WAR AGAINST THE UKRAINIAN STATE

The purpose of this article is to clarify the essence and scope of Ukraine's victory in the context of Russia's total war against the Ukrainian state. A system-situational approach, historical analysis, and modeling methods were employed to address the research questions. The article demonstrates that a military victory in the physical realm becomes less significant than its representation in virtual space, where it can be shaped through propaganda in the public consciousness. The factors influencing the dynamics of scientific and official discourse regarding perceptions of victory in modern warfare are identified. It has been established that the essence of Ukraine's victory in the total Russian-Ukrainian war is as follows: in the military context – the defeat of Russia and Ukraine's achievement of the political and military-strategic objectives of its national and just war; in the political context – subordinating Russia to the goals and objectives of Ukraine's diplomatic and strategic efforts aimed at restoring its territorial integrity; and in the praxeological context – Ukraine's achievement of a peace superior to that which existed before the war. Ukraine's victory in the total Russian-Ukrainian war, in substantive terms, consists of: in the military context – the defeat of Russian armed forces and the cessation of hostilities within Ukraine's borders; in the political context – the restoration of Ukraine’s territorial integrity; and in the international political context – achieving victory with guarantees of state sovereignty, territorial integrity, and the inviolability of borders from NATO and the EU, as well as securing peace without Russia's geopolitical dominance, a peace based on the rule of international law rather than the rule of force.

Joint self-supervised and supervised contrastive learning for multimodal MRI data: Towards predicting abnormal neurodevelopment

The integration of different imaging modalities, such as structural, diffusion tensor, and functional magnetic resonance imaging, with deep learning models has yielded promising outcomes in discerning phenotypic characteristics and enhancing disease diagnosis. The development of such a technique hinges on the efficient fusion of heterogeneous multimodal features, which initially reside within distinct representation spaces. Naively fusing the multimodal features does not adequately capture the complementary information and could even produce redundancy. In this work, we present a novel joint self-supervised and supervised contrastive learning method to learn the robust latent feature representation from multimodal MRI data, allowing the projection of heterogeneous features into a shared common space, and thereby amalgamating both complementary and analogous information across various modalities and among similar subjects. We performed a comparative analysis between our proposed method and alternative deep multimodal learning approaches. Through extensive experiments on two independent datasets, the results demonstrated that our method is significantly superior to several other deep multimodal learning methods in predicting abnormal neurodevelopment. Our method has the capability to facilitate computer-aided diagnosis within clinical practice, harnessing the power of multimodal data. The source code of the proposed model is publicly accessible on GitHub: https://github.com/leonzyzy/Contrastive-Network.

Study on Self-Learning System for Milling Chatter Feature Based on Hybrid Preprocessing Model and Improved Convolutional Clustering

Milling chatter is a self-excited vibration phenomenon during the cutting process, typically occurring in machining operations with lower stiffness. It significantly impacts the machining precision and surface morphology of the workpiece. To summarize the vibration signal features under operating conditions and enhance the precision of the intelligent recognition model for milling chatter features, this paper proposes an adaptive learning model for chatter feature recognition based on the improved convolutional clustering algorithm. This study uses wavelet analysis to extract chatter features and improved convolutional clustering to identify chatter and stable cutting signals. This method selected appropriate mapping techniques based on differences in chatter signals under various conditions and uses convolution for feature extraction. Improve convolutional clustering evaluates sample similarity through energy transfer, less influenced by signal space structure compared to Euclidean or Mahalanobis distances. This method provides a unified criterion for evaluating different signal feature representation spaces, thereby improving the accuracy and efficiency of chatter sample recognition. The recognition accuracy of chatter phenomena at different spindle speed ranges reaches $95$ % and $96.3$ % respectively. The results show that this method can effectively distinguish between chatter signals and stable cutting signals.

Parametric Optimization for Fully Fuzzy Linear Programming Problems with Triangular Fuzzy Numbers

This paper presents a new approach for solving FFLP problems using a double parametric form (DPF), which is critical in decision-making scenarios characterized by uncertainty and imprecision. Traditional linear programming methods often fall short in handling the inherent vagueness in real-world problems. To address this gap, an innovative method has been proposed which incorporates fuzzy logic to model the uncertain parameters as TFNs, allowing for a more realistic and flexible representation of the problem space. The proposed method stands out due to its integration of fuzzy arithmetic into the optimization process, enabling the handling of fuzzy constraints and objectives directly. Unlike conventional techniques that rely on crisp approximations or the defuzzification process, the proposed approach maintains the fuzziness throughout the computation, ensuring that the solutions retain their fuzzy characteristics and better reflect the uncertainties present in the input data. In summary, the proposed method has the ability to directly incorporate fuzzy parameters into the optimization framework, providing a more comprehensive solution to FFLP problems. The main findings of this study underscore the method’s effectiveness and its potential for broader application in various fields where decision-making under uncertainty is crucial.

Efficient Neural Decoding Based on Multimodal Training.

Neural decoding methods are often limited by the performance of brain encoders, which map complex brain signals into a latent representation space of perception information. These brain encoders are constrained by the limited amount of paired brain and stimuli data available for training, making it challenging to learn rich neural representations. To address this limitation, we present a novel multimodal training approach using paired image and functional magnetic resonance imaging (fMRI) data to establish a brain masked autoencoder that learns the interactions between images and brain activities. Subsequently, we employ a diffusion model conditioned on brain data to decode realistic images. Our method achieves high-quality decoding results in semantic contents and low-level visual attributes, outperforming previous methods both qualitatively and quantitatively, while maintaining computational efficiency. Additionally, our method is applied to decode artificial patterns across region of interests (ROIs) to explore their functional properties. We not only validate existing knowledge concerning ROIs but also unveil new insights, such as the synergy between early visual cortex and higher-level scene ROIs, as well as the competition within the higher-level scene ROIs. These findings provide valuable insights for future directions in the field of neural decoding.

Classification of the difficulty of a climbing route using the transformation of representation spaces and cascading classifier ensemble (CCE)

Climbing has gained popularity in recent years and encompasses various disciplines, among which bouldering stands out as one of the most well-known. Determining the difficulty of a bouldering route is a challenging task due to the varying combinations of handholds. Accurate assessment of the difficulty of a route is important, as it allows a climber’s ability to be measured against that of other climbers. Furthermore, climbers must gradually progress from less to more difficult routes to improve their skills effectively. The difficulty rating of a route poses a complex problem, and previous research has explored machine learning techniques, including deep learning, to address it. We propose a novel approach based on a series of transformations in the representation space that incorporates expert knowledge to improve the classification of climbing route difficulty. We introduce a cascade classifier ensemble (CCE) model specifically designed for levelling classification. Compared to similar works, this model also incorporates explainable artificial intelligence (XAI) mechanisms. With the CEE model, the relevance of specific handholds in the transition from one level of difficulty to the next is obtained. Finally, we illustrate how our proposed model not only achieves good results but also empowers climbers to identify the most crucial sections in each level of difficulty.

What Is Faster than Where in Vocal Emotional Perception.

Voices carry a vast amount of information about speakers (e.g., emotional state; spatial location). Neuroimaging studies postulate that spatial ("where") and emotional ("what") cues are processed by partially independent processing streams. Although behavioral evidence reveals interactions between emotion and space, the temporal dynamics of these processes in the brain and its modulation by attention remain unknown. We investigated whether and how spatial and emotional features interact during voice processing as a function of attention focus. Spatialized nonverbal vocalizations differing in valence (neutral, amusement, anger) were presented at different locations around the head, whereas listeners discriminated either the spatial location or emotional quality of the voice. Neural activity was measured with ERPs of the EEG. Affective ratings were collected at the end of the EEG session. Emotional vocalizations elicited decreased N1 but increased P2 and late positive potential amplitudes. Interactions of space and emotion occurred at the salience detection stage: neutral vocalizations presented at right (vs. left) locations elicited increased P2 amplitudes, but no such differences were observed for emotional vocalizations. When task instructions involved emotion categorization, the P2 was increased for vocalizations presented at front (vs. back) locations. Behaviorally, only valence and arousal ratings showed emotion-space interactions. These findings suggest that emotional representations are activated earlier than spatial representations in voice processing. The perceptual prioritization of emotional cues occurred irrespective of task instructions but was not paralleled by an augmented stimulus representation in space. These findings support the differential responding to emotional information by auditory processing pathways.

Designing Connectivity-Guaranteed Porous Metamaterial Units Using Generative Graph Neural Networks

Abstract Designing 3D porous metamaterial units while ensuring complete connectivity of both solid and pore phases presents a significant challenge. This complete connectivity is crucial for manufacturability and structure-fluid interaction applications (e.g., fluid-filled lattices). In this study, we propose a generative graph neural network-based framework for designing the porous metamaterial units with the constraint of complete connectivity. First, we propose a graph-based metamaterial unit generation approach to generate porous metamaterial samples with complete connectivity in both solid and pore phases. Second, we establish and evaluate three distinct variational graph autoencoder (VGAE)-based generative models to assess their effectiveness in generating an accurate latent space representation of metamaterial structures. By choosing the model with the highest reconstruction accuracy, the property-driven design search is conducted to obtain novel metamaterial unit designs with the targeted properties. A case study on designing liquid-filled metamaterials for thermal conductivity properties is carried out. The effectiveness of the proposed graph neural network-based design framework is evaluated by comparing the performances of the obtained designs with those of known designs in the metamaterial database. Merits and shortcomings of the proposed framework are also discussed.