Information Compression Research Articles

A compressibility account of the color-sharing bonus in working memory.

It has been established that objects sharing color in a visual display can boost working memory. The capacity to encode singletons particularly benefits from the repetition of colors encoded as perceptual groups. We manipulated the algorithmic complexity of visual displays to test whether compressibility of information could account for the color-sharing bonus. This study used a free recall working memory task in which the participants were shown displays of 2 to 8 color items. We examined the influence of set size, complexity, number of same-color clusters and amount of color redundancy. The results showed that the probability of correct recall of the pattern and the proportion of similarity between the pattern and the response decreased with an increase of each manipulated variable, except for color redundancy in terms of probability of correct recall. The model performance of complexity did not differ from that of clusters, but complexity was found more accurate than either set size or color redundancy. The results also showed that similar items were more often recalled adjacently, and complexity correlated strongly with the number of extra color repetitions in the response, suggesting that more complex patterns encouraged the use of information compression. Moreover, color repetitions were more often recalled first and the probability of correct recall for singletons and sub-patterns could be predicted by the compressibility measure. We discuss the potential advantage of using compressibility measures to capture the effects of regularities in visual patterns, in particular to refine analysis of the color-sharing bonus.

Linguistic laws of brevity: conformity in Indri indri.

Vocal and gestural sequences of several primates have been found to conform to two general principles of information compression: the compensation between the duration of a construct and that of its components (Menzerath-Altmann law) and an inverse relationship between signal duration and its occurrence (Zipf's law of abbreviation). Even though Zipf's law of brevity has been proposed as a universal in animal communication, evidence on non-human primate vocal behavior conformity to linguistic laws is still debated, and information on strepsirrhine primates is lacking. We analyzed the vocal behavior of the unique singing lemur species (Indri indri) to assess whether the song of the species shows evidence for compression. As roars have a chaotic structure that impedes the recognition of each individual utterance, and long notes are usually given by males, we focused on the core part of the song (i.e., the descending phrases, composed of two-six units). Our results indicate that indris' songs conform to Zipf's and Menzerath-Altmann linguistic laws. Indeed, shorter phrases are more likely to be included in the song, and units' duration decrease at the increase of the size of the phrases. We also found that, despite a sexual dimorphism in the duration of both units and phrases, these laws characterize sequences of both males and females. Overall, we provide the first evidence for a trade-off between signal duration and occurrence in the vocal behavior of a strepsirrhine species, suggesting that selective pressures for vocal compression are more ancestral than previously assumed within primates.

Color Conversion in Deep Autoencoders

Abstract While RGB is the status quo in machine vision, other color spaces offer higher utility in distinct visual tasks. Here, the authors have investigated the impact of color spaces on the encoding capacity of a visual system that is subject to information compression, specifically variational autoencoders (VAEs) with a bottleneck constraint. To this end, they propose a framework—color conversion—that allows a fair comparison of color spaces. They systematically investigated several ColourConvNets, i.e. VAEs with different input‐output color spaces, e.g. from RGB to CIE L∗a∗b∗ (in total five color spaces were examined). Their evaluations demonstrate that, in comparison to the baseline network (whose input and output are RGB), ColourConvNets with a color-opponent output space produce higher quality images. This is also evident quantitatively: (i) in pixel-wise low-level metrics such as color difference (ΔE), peak signal-to-noise ratio (PSNR) and structural similarity index measure (SSIM); and (ii) in high-level visual tasks such as image classification (on ImageNet dataset) and scene segmentation (on COCO dataset) where the global content of reconstruction matters. These findings offer a promising line of investigation for other applications of VAEs. Furthermore, they provide empirical evidence on the benefits of color-opponent representation in a complex visual system and why it might have emerged in the human brain.

Three new lengths for cyclic Legendre pairs

Introduction: It is conjectured that the cyclic Legendre pairs of odd lengths >1 always exist. Such a pair consists of two functions a, b: G→Z, whose values are +1 or −1, and whose periodic autocorrelation function adds up to the constant value −2 (except at the origin). Here G is a finite cyclic group and Z is the ring of integers. These conditions are fundamental and the closely related structure of Hadamard matrices having a two circulant core and double border is incompletely described in literature, which makes its study especially relevant. Purpose: To describe the two-border two-circulant-core construction for Legendre pairs having three new lengths. Results: To construct new Legendre pairs we use the subsets X={x∈G: a(x)=–1} and Y={x∈G: b(x)=–1} of G. There are 20 odd integers v less than 200 for which the existence of Legendre pairs of length v is undecided. The smallest among them is v=77. We have constructed Legendre pairs of lengths 91, 93 and 123 reducing thereby the number of undecided cases to 17. In the last section of the paper we list some new examples of cyclic Legendre pairs for lengths v≤123. Practical relevance: Hadamard matrices are used extensively in the problems of error-free coding, and compression and masking of video information. Programs for search of Hadamard matrices and a library of constructed matrices are used in the mathematical network “mathscinet.ru” together with executable on-line algorithms

The Potential of the SP System in Machine Learning and Data Analysis for Image Processing

This paper aims to describe how pattern recognition and scene analysis may with advantage be viewed from the perspective of the SP system (meaning the SP theory of intelligence and its realisation in the SP computer model (SPCM), both described in an appendix), and the strengths and potential of the system in those areas. In keeping with evidence for the importance of information compression (IC) in human learning, perception, and cognition, IC is central in the structure and workings of the SPCM. Most of that IC is achieved via the powerful concept of SP-multiple-alignment, which is largely responsible for the AI-related versatility of the system. With examples from the SPCM, the paper describes: how syntactic parsing and pattern recognition may be achieved, with corresponding potential for visual parsing and scene analysis; how those processes are robust in the face of errors in input data; how in keeping with what people do, the SP system can “see” things in its data that are not objectively present; the system can recognise things at multiple levels of abstraction and via part-whole hierarchies, and via an integration of the two; the system also has potential for the creation of a 3D construct from pictures of a 3D object from different viewpoints, and for the recognition of 3D entities.

Natural Product Synthesis through the Lens of Informatics.

Retrosynthetic analysis emerged in the 1960s as a teaching tool with profound implications. Its educational value can be appreciated by a glance at total synthesis manuscripts over 50 years later, most of which contain a retrosynthesis on page one. Its vision extended to computer language-a pioneering idea in the 20th century that continues to expand the frontiers today. The same principles that guide a student to evaluate, expand, and refine a series of bond dissections can be programmed, so that computer assistance can perform the same tasks but at faster speeds.The slow step in the synthesis of complex structures, however, is seldom route design. Compression of molecular information into close proximity (Cm/Å3) requires exploration and empiricism, a close connection between theory and experiment. Here, retrosynthetic analysis guides the choice of experiment, so that the most simplifying-but often least assured-disconnection is prioritized: a high-risk, high reward strategy. The reimagining of total synthesis in a future era of retrosynthetic software may involve, counterintuitively, target design, as discussed here.Compared to the 1960s, retrosynthetic analysis in the 21st century finds itself among computers of unimaginable power and a biology that is increasingly molecular. Put together, the logic of retrosynthesis, the insight of structural biology, and the predictions of computation have inspired us to imagine an integration of the three. The synthetic target is treated as dynamic-a constellation of related structures-in order to find the nearest congener with the closest affinity but the shortest synthetic route. Such an approach merges synthetic design with structural design toward the goal of improved access for improved function.In this Account, we detail the evolution of our program from its inception in traditional natural product (NP) total synthesis to its current expression through the lens of chemical informatics: a view of NPs as aggregates of molecular parameters that define single points in a chemical space. Early work on synthesis and biological annotation of apparent metal pool binders and nonselective covalent electrophiles (asmarine alkaloids, isocyanoterpenes, Nuphar dimers) gave way to NPs with well-defined protein targets. The plant metabolite salvinorin A (SalA) potently and selectively agonizes the κ-opioid receptor (KOR), rapidly penetrates the brain, and represents an important lead for next-generation analgesics and antipruritics. To synthesize and diversify this lead, we adopted what we now call a dynamic approach. Deletion of a central methyl group stabilized the SalA scaffold, opened quick synthetic access, and retained high potency and selectivity. The generality of this idea was then tested against another neuroactive class. As an alternative hypothesis to TrkB channels, we proposed that the so-called "neurotrophic" Illicium terpenes may bind to γ-aminobutyric acid (GABA)-gated ion channels to cause weak, chronic excitation. Syntheses of (-)-jiadifenolide, 3,6-dideoxy-10-hydroxypseudoanisatin, (-)-11-O-debenzoyltashironin, (-)-bilobalide, and (-)-picrotoxinin (PXN) allowed this hypothesis to be probed more broadly. Feedback from protein structure and synthetic reconnaissance led to a dynamic retrosynthesis of PXN and the identification of 5MePXN, a moderate GABAAR antagonist with greater aqueous stability available in eight steps from dimethylcarvone. We expect this dynamic approach to synthetic target analysis to become more feasible in the coming years and hope the next generation of scientists finds this approach helpful to address problems at the frontier of chemistry and biology.

Particle Swarm Optimization Based on a Novel Evaluation of Diversity

In this paper, we propose a particle swarm optimization variant based on a novel evaluation of diversity (PSO-ED). By a novel encoding of the sub-space of the search space and the hash table technique, the diversity of the swarm can be evaluated efficiently without any information compression. This paper proposes a notion of exploration degree based on the diversity of the swarm in the exploration, exploitation, and convergence states to characterize the degree of demand for the dispersion of the swarm. Further, a disturbance update mode is proposed to help the particles jump to the promising regions while reducing the cost of function evaluations for poor particles. The effectiveness of PSO-ED is validated on the CEC2015 test suite by comparison with seven popular PSO variants out of 12 benchmark functions; PSO-ED achieves six best results for both 10-D and 30-D.

Learning Efficient Binarized Object Detectors With Information Compression.

In this paper, we propose a binarized neural network learning method (BiDet) for efficient object detection. Conventional network binarization methods directly quantize the weights and activations in one-stage or two-stage detectors with constrained representational capacity, so that the information redundancy in the networks causes numerous false positives and degrades the performance significantly. On the contrary, our BiDet fully utilizes the representational capacity of the binary neural networks by redundancy removal, through which the detection precision is enhanced with alleviated false positives. Specifically, we generalize the information bottleneck (IB) principle to object detection, where the amount of information in the high-level feature maps is constrained and the mutual information between the feature maps and object detection is maximized. Meanwhile, we learn sparse object priors so that the posteriors are concentrated on informative detection prediction with false positive elimination. Since BiDet employs a fixed IB trade-off to balance the total and relative information contained in the high-level feature maps, the information compression leads to ineffective utilization of the network capacity or insufficient redundancy removal for input in different complexity. To address this, we further present binary neural networks with automatic information compression (AutoBiDet) to automatically adjust the IB trade-off for each input according to the complexity. Moreover, we further propose the class-aware sparse object priors by assigning different sparsity to objects in various classes, so that the false positives are alleviated more effectively without recall decrease. Extensive experiments on the PASCAL VOC and COCO datasets show that our BiDet and AutoBiDet outperform the state-of-the-art binarized object detectors by a sizable margin.

SPECTRAL ANALYSIS AND FORECASTING OF THE 25TH SOLAR CYCLES

In the present work, we propose an extrapolation method, developed on the basis of spectral analysis, digital filtering, and the principle of demodulation of a complex signal, for predicting the beginning of cycle 25 of solar activity. The Wolf number and other measured characteristics of solar activity have a very complex spectral composition. The Sun, by the nature of its radiation, contributes a significant stochastic component to the observational data. The experimental data are known only up to the present, and the prediction is about bridging the gap in our data set. Mathematically, the prediction problem boils down to extrapolation of discontinuous functions, which leads to a Gibbs phenomenon that occurs at the point of discontinuity and makes prediction into the future impossible. To overcome this discontinuity, additional physical models describing a continuous process are most often used. This paper uses only the Wolf series of numbers from 1818 to 2020. The authors developed an original forecasting technique using Fourier series, digital filtering and representation of the complex process as modulated and subsequent demodulation. As a result of decomposing the complex signal by Fourier series into separate components, the spectral ranges characteristic of the Wolf number were singled out. Taylor's series was used for construction of prediction or extrapolation algorithms. The extraction of spectral ranges, characteristic for the investigated process, is carried out by means of sequential digital filtering methods and information compression in accordance with the cut-off frequency of the digital filter. For example, when selecting eleven-year cycles of solar activity, we have to compress the information by a factor of 160. With such a processing scheme, the forecasting starts with the ultralow-frequency component with a period of more than 11 years, successively moving to the ranges of higher frequencies. The use of spectral analysis and Chebyshev filtering showed the possibility to predict the low-frequency component for the full cycle period. The eleven-year component forecast obtained by the authors is in good agreement with the data of the Brussels Royal Center.

Multi-Scale and Multi-Channel Networks for CSI Feedback in Massive MIMO System

In the frequency division duplex (FDD) mode of the massive MIMO system, the system needs to perform coding through channel state information (CSI) to obtain performance gains. However, the number of antennas of the base station has been greatly increased, resulting in a rapid increase in the overhead for the user terminal to feedback CSI to the base station. In this article, we propose a method based on multi-task CNN to achieve compression and reconstruction of channel state information through a multi-scale and multi-channel convolutional neural network. We also introduce a dynamic learning rate model to improve the accuracy of channel state information reconstruction. The simulation results show that compared with the original CsiNet and other work, the proposed CSI feedback network has better reconstruction performance.

Method for building a model of the functioning of the on-board equipment of spacecrafts based on a classification scheme of interactions of subsystems

In this paper, a promising approach to solving the problem of adaptive-distributed control of the technical condition of onboard equipment of spacecraft based on the classification scheme of the mutual influence of subsystems is determined. Purpose is to improve the efficiency of autonomous monitoring of the technical condition of spacecraft on-board equipment. The proposed method for constructing a model of the functioning of onboard equipment of spacecraft made it possible to obtain a classification scheme for the mutual influence of subsystems of onboard equipment, the use of which as a vehicle control model provides semantic compression of telemetric information. The idea of creating a special software for advanced on-board control systems with the function of adaptive-distributed control of technical condition is proposed. The possibility of using the methodology is not excluded when constructing a model of the operation process of on-board equipment for objects of various purposes, including as an element of an artificial intelligence system when solving the problem of monitoring a technical condition.

Configurational complexity of nonautonomous discrete one-soliton and rogue waves in Ablowitz-Ladik-Hirota waveguide

We compute the configurational complexity (CC) for discrete soliton and rogue waves traveling along an Ablowitz-Ladik-Hirota (ALH) waveguide and modeled by a discrete nonlinear Schrödinger equation. We show that for a specific range of the soliton transverse direction κ propagating along the parametric time ζ(t), CC reaches an evolving series of global minima. These minima represent maximum compressibility of information in the momentum modes along the Ablowitz-Ladik-Hirota waveguide. Computing the CC for rogue waves as a function of background amplitude modulation μ, we show that it displays two essential features: a maximum representing the optimal value for the rogue wave inception (the “gradient catastrophe”) and saturation representing the rogue wave dispersion into constituent wave modes. We show that saturation is achieved earlier for higher values of modulation amplitude as the discrete rogue wave evolves along time ζ(t).

Speaker stance and evaluative -ly adverbs in the Modern English period

Present-day English makes major use of adverbs for the expression of speaker stance, following what has been described as the adverbialization of speaker attitude. Two adverbial constructions have particular stance-expressing functions: those with an adverb that modifies a clause or sentence (SAdv) and those with an adverb that modifies an adjective phrase (AdjAdv). The emergence of modal and evaluative SAdvs from verb-phrase-modifying adverbs (VPAdvs) via inferencing and scope expansion has been well documented. There is a large literature on the diachronic development of Present-day English intensifiers such as absolutely or really. Much less attention has been paid to the evaluative function of many AdjAdvs. This paper focuses on the relationship between extra-clausal stance -ly adverbs of the SAdv construction and their intra-clausal counterparts of the AdjAdv construction. It is argued that -ly adverbial modification of adjectives cannot be reduced to intensification. It has become a favoured site for the expression of speaker evaluation, contributing to increase in information compression. The evaluative AdjAdv and SAdv constructions are seen to share a semantic schema, and by hypothesis belong to a single more abstract constructional schema. While earlier types appear to have evolved from circumstance adverbs, later types may have been coined by analogy in a cycle of increasing productivity.

DEVELOPMENT OF AN ADVANCED METHOD OF VIDEO INFORMATION RESOURCE COMPRESSION IN NAVIGATION AND TRAFFIC CONTROL SYSTEMS

The Earth's aerospace monitoring (ASM) systems use state-of-the-art integrated information technologies that include radio-based detection and surveillance systems using telecommunications. One of the main tasks of ASM systems is to increase the efficiency of decision-making necessary for the timely prevention, detection, localization and elimination of crisis situations and their probable consequences. Modern conditions impose stricter requirements for efficiency, reliability and quality of the provided video data. To ensure compliance with the requirements, it is necessary to provide the appropriate capabilities of the onboard equipment. On the basis of the existing information and communication systems it is necessary to carry out: continuous or periodic assessment of a condition of objects of supervision and control; continuous (operational) collection, reception, transmission, processing, analysis and display of information resources. It is proposed to use UAVs (unmanned aerial vehicles) as a means to perform ASM tasks. The time of organizing communication sessions and delivery of information should vary from a few seconds to 2.5 hours. Untimely processing and delivery of a specific information resource in the management process leads to its obsolescence or loss of relevance, which contributes to erroneous decisions. One way to reduce time is to encode the data. To do this, it is proposed to use video compression algorithms. However, based on the analysis of the possibility of modern methods of video information compression, taking into account the specifics of the onboard equipment of the UAV, the coding problem is not completely solved. The research results show the expediency of using an improved method of video information compression to reduce the computing resources of the software and hardware complex of the onboard UAV equipment and to ensure the requirements for efficiency and reliability of data in modern threats to ASM systems as a whole.

Spectrum projection with a bandgap-gradient perovskite cell for colour perception

Optoelectronic devices for light or spectral signal detection are desired for use in a wide range of applications, including sensing, imaging, optical communications, and in situ characterization. However, existing photodetectors indicate only light intensities, whereas multiphotosensor spectrometers require at least a chip-level assembly and can generate redundant signals for applications that do not need detailed spectral information. Inspired by human visual and psychological light perceptions, the compression of spectral information into representative intensities and colours may simplify spectrum processing at the device level. Here, we propose a concept of spectrum projection using a bandgap-gradient semiconductor cell for intensity and colour perception. Bandgap-gradient perovskites, prepared by a halide-exchanging method via dipping in a solution, are developed as the photoactive layer of the cell. The fabricated cell produces two output signals: one shows linear responses to both photon energy and flux, while the other depends on only photon flux. Thus, by combining the two signals, the single device can project the monochromatic and broadband spectra into the total photon fluxes and average photon energies (i.e., intensities and hues), which are in good agreement with those obtained from a commercial photodetector and spectrometer. Under changing illumination in real time, the prepared device can instantaneously provide intensity and hue results. In addition, the flexibility and chemical/bio-sensing of the device via colour comparison are demonstrated. Therefore, this work shows a human visual-like method of spectrum projection and colour perception based on a single device, providing a paradigm for high-efficiency spectrum-processing applications.

Service Function Path Provisioning With Topology Aggregation in Multi-Domain Optical Networks

Traffic flows are often processed by a chain of Service Functions (SFs) (known as Service Function Chaining (SFC)) to satisfy service requirements. The deployed path for a SFC is called Service Function Path (SFP). SFs can be virtualized and migrated to datacenters, thanks to the evolution of Software Defined Network (SDN) and Network Function Virtualization (NFV). In such a scenario, provisioning of paths (i.e., SFPs) between virtualized network functions is an important problem. SFP provisioning becomes more complex in a multi-domain network topology. `Topology aggregation' helps to create a single-domain view of such a network by abstracting multi-domain networks. However, traditional `topology aggregation' methods are unable to abstract SF resources properly, which is required for SFP provisioning. In this paper, we propose an SFC-Oriented Topology Aggregation (SOTA) method to enable abstraction for SFs in multi-domain optical networks. This study explores the node and the link aggregation degree to evaluate information compression during the `Topology aggregation' process. Additionally, we also propose a new data structure named wheel matrix and related operations to store routing information in the aggregated topology. Based on SOTA, we propose two cross-domain SFP provisioning algorithms named Ordered Anchor Selection (OAS) and ${k}$ -paths OAS (K-OAS), and a benchmark named Global OAS (GOAS). Simulation results show that SOTA could aggregate large-scale multi-domain optical networks into a small network that contains only 6.9% of the nodes and 10.1% of the links. Both OAS and K-OAS can calculate SFPs efficiently and reduce blocking probability up to 52.10% compared to the benchmark.

Increasing the efficiency of information transmission in communication channels

<span lang="EN-US">This paper discusses compression methods focused on data transmission over communication channels. The characteristics of different algorithms for different types of incoming data are analyzed. The purpose of this study is to evaluate the speed of operation of each of the compression algorithms for different types of information and different compression parameters, on the basis of the obtained results to make recommendations for the application of compression methods in systems critical to the performance of the algorithm. Based on the results of the analysis, the methods of compression that can be used in communication channels are selected: LZW, LZH, Vitter and matrix. The practical research of the selected methods on different information flows (text, graphics, measurement data, combined data) was carried out, their comparative analysis was performed. Research has highlighted compression methods that give the most optimal results in each case. Comparative evaluation of algorithms for different parameters is made, the possibility of data compression implementation in systems running in real time is analyzed. Based on the results of the study, recommendations are made for the application of particular compression methods in specific conditions.</span>

Dynamic Compression Ratio Selection for Edge Inference Systems With Hard Deadlines

Implementing machine learning algorithms on the Internet-of-Things (IoT) devices has become essential for emerging applications, such as autonomous driving and environment monitoring. But the limitations of computation capability and energy consumption make it difficult to run complex machine learning algorithms on IoT devices, especially when the latency deadline exists. One solution is to offload the computation-intensive tasks to the edge server. However, the wireless uploading of the raw data is time consuming and may lead to deadline violation. To reduce the communication cost, lossy data compression can be exploited for inference tasks but may bring more erroneous inference results. In this article, we propose a dynamic compression ratio selection scheme for edge inference system with hard deadlines. The key idea is to balance the tradeoff between the communication cost and inference accuracy. By dynamically selecting the optimal compression ratio with the remaining deadline budgets for queued tasks, more tasks can be timely completed with correct inference under limited communication resources. Furthermore, information augmentation that retransmits less compressed data of task with erroneous inference, is proposed to enhance the accuracy performance. While it is often hard to know the correctness of inference, we use uncertainty to estimate the confidence of the inference and based on that, jointly optimize the information augmentation and compression ratio selection. Finally, considering the wireless transmission errors, we further design a retransmission scheme to reduce performance degradation due to packet losses. The simulation results show the performance of the proposed schemes under different deadlines and task arrival rates.

Conference matrices from Legendre C-pairs

Introduction: There are just a few known methods for the construction of symmetric C-matrices, due to the lack of a universal structure for them. This obstruction is fundamental, in addition, the structure of C-matrices with a double border is incompletely described in literature, which makes its study especially relevant. The purpose: To describe the two-border two-circulant construction in detail with the proposal of the concept of C-pairs Legendre. Results: The paper deals with C-matrices of order n=2v+2 with two borders and extends the so called generalized Legendre pairs, v odd, to a wider class of Legendre C-pairs with even and odd v, defined on a finite abelian group G of order v. Such a pair consists of two functions a, b: G→Z, whose values are +1 or −1 except that a(e)=0, where e is the identity element of G and Z is the ring of integers. To characterize the Legendre C-pairs we use the subsets X={xÎG: a(x)=–1} and Y={xÎG: b(x)=–1} of G. We show that a(x−1)=(−1)v a(x) for all x. For odd v we show that X and Y form a difference family, which is not true for even v. These difference families are precisely the so called Szekeres difference sets, used originally for the construction of skew-Hadamard matrices. We introduce the subclass of the special Legendre C-pairs and prove that they exist whenever 2v+1 is a prime power. In the last two sections of the paper we list examples of special cyclic Legendre C-pairs for lengths v<70. Practical relevance: C-matrices are used extensively in the problems of error-free coding, compression and masking of video information. Programs for search of conference matrices and a library of constructed matrices are used in the mathematical network “mathscinet.ru” together with executable on-line algorithms.

Causal analysis of competing atomistic mechanisms in ferroelectric materials from high-resolution scanning transmission electron microscopy data

Machine learning has emerged as a powerful tool for the analysis of mesoscopic and atomically resolved images and spectroscopy in electron and scanning probe microscopy, with the applications ranging from feature extraction to information compression and elucidation of relevant order parameters to inversion of imaging data to reconstruct structural models. However, the fundamental limitation of machine learning methods is their correlative nature, leading to extreme susceptibility to confounding factors. Here, we implement the workflow for causal analysis of structural scanning transmission electron microscopy (STEM) data and explore the interplay between physical and chemical effects in a ferroelectric perovskite across the ferroelectric–antiferroelectric phase transitions. The combinatorial library of the Sm-doped BiFeO3 is grown to cover the composition range from pure ferroelectric BFO to orthorhombic 20% Sm-doped BFO. Atomically resolved STEM images are acquired for selected compositions and are used to create a set of local compositional, structural, and polarization field descriptors. The information-geometric causal inference (IGCI) and additive noise model (ANM) analysis are used to establish the pairwise causal directions between the descriptors, ordering the data set in the causal direction. The causal chain for IGCI and ANM across the composition is compared and suggests the presence of common causal mechanisms across the composition series. Ultimately, we believe that the causal analysis of the multimodal data will allow exploring the causal links between multiple competing mechanisms that control the emergence of unique functionalities of morphotropic materials and ferroelectric relaxors.