Cross-domain structured mapping in machine learning processing

Abstract

This disclosure relates to cross-domain structured mapping in machine learning processing. A method using a computational device to correlate two or more corpora of dissimilar data includes receiving input data from each of the two or more corpora of dissimilar data. The computational device computes a pass-through of each of the input data to two or more encoder-decoder models. The computational device also obtains a prediction for identity mapping for each different knowledge domain from each of the two or more encoder-decoder models. The computational device additionally computes a distributional distance metric as the output of each of the low-dimensional embedding vector representations from each of the two or more encoder-decoder models. The computational device also computes a function based on the predictions from each of the two or more encoder-decoder models and each of the distributional distance metrics. The computational device further updates the two or more encoder-decoder models.

Description

Technical Field

[0001] The field of this invention relates to machine learning models and systems for cross-domain structured mapping of various domain entity file types. Background Technology

[0002] Many problems in industry, science, and research can be solved by inspiration from other orthogonal domains with data from individual domains available via the internet, dedicated networks, and collections (e.g., documents, images, videos), thus allowing for cross-domain innovation in the design of solutions to problems in domains that may have similar concepts but in different contexts. However, the scalability of cross-domain innovation capabilities requires experts in each domain and strong collaboration in identifying these similar patterns across different domains, which is both time-consuming and expensive. Summary of the Invention

[0003] The embodiments relate to cross-domain structured mapping machine learning models and systems. One embodiment provides a method for using a computational device to correlate two or more corpora of dissimilar data, the method comprising receiving input data from each of the two or more corpora of dissimilar data. The computational device computes a pass-through of each of the input data to two or more encoder-decoder models. The computational device also obtains a prediction of identity mapping for each distinct knowledge domain from each of the two or more encoder-decoder models. The computational device additionally computes a distributional distance metric as the output of each of the low-dimensional embedding vector representations from each of the two or more encoder-decoder models. The computational device also computes a function based on the predictions from each of the two or more encoder-decoder models and each of the distributional distance metrics. The computational device further updates the two or more encoder-decoder models. The embodiments significantly improve the efficiency of researchers in simplifying cross-disciplinary source discovery and matching without significant knowledge of other external domains. Some features contribute to the advantages of discovering new assets, extracting and associating different components of documents, and mapping them to relevant proposals and other products. Several other features help to identify the merits of researchers’ work in new applications in new fields and encourage the efficient use of their work and assets by reusing them in different regions.

[0004] It may include one or more of the following features. In some embodiments, the method may further include having a computing device compute a corresponding reconstruction loss for each of two or more encoder-decoder models using corresponding predictions and input data from each of two or more corpora of dissimilar data. The computing device may also include extracting low-dimensional embedding vectors of the input data representation from each of the two or more encoder-decoder models.

[0005] In some embodiments, the method may further include: the distribution distance metric is a pairwise average relative survival time (MRLT) distribution distance metric, and the function is a joint loss function.

[0006] In one or more embodiments, the method may further include having a computing device compute the gradient of the loss from the joint loss function for the model parameters of each of two or more encoder-decoder models.

[0007] In some embodiments, the method may further include: initializing the weights of each of the two or more encoder-decoder models by the computing device. The computing device further performs preprocessing, transformation, and extraction of the input data into fixed-dimensional feature vectors. The computing device further performs feedforward processing to feedforward each in-domain sample of the input data to each corresponding model of the two or more encoder-decoder models. The computing device further uses each corresponding model of the two or more encoder-decoder models to generate a corresponding output prediction for each in-domain sample of the input data. Given the in-domain samples of the input data and the corresponding output predictions, the computing device also computes the corresponding loss value of the joint loss function for each of the two or more encoder-decoder models.

[0008] In one or more embodiments, the method may include computing a pairwise MRLT distribution distance metric based on a first relative survival time (RLT) matrix and a second RLT matrix between samples within each domain of the input data and based on the distribution distance between the first RLT matrix and the second RLT matrix.

[0009] In some embodiments, the method may further include computing a pairwise MRLT distribution distance metric based on a first RLT matrix and a second RLT matrix between samples within each domain of the input data and based on a squared loss function between the outputs of the first RLT matrix and the outputs of the second RLT matrix.

[0010] In one or more embodiments, the method may include computing a pairwise MRLT distribution distance metric based on an RLT matrix and a second RLT matrix between samples within each domain of the input data and based on Wasserstein distance determination of the distribution using the first RLT matrix and the second RLT matrix.

[0011] In some embodiments, the method may include: two or more corpora containing dissimilar data comprising text, images, audio, and other data sources in different knowledge domains.

[0012] These and other features, aspects and advantages of the embodiments of the present invention will be understood with reference to the following description, the appended claims and the accompanying drawings. Attached Figure Description

[0013] Figure 1 An example is shown of using topological representations to compare different representations of a domain, according to one embodiment, which uses a defined structure to jointly learn the appropriate mapping between these representations;

[0014] Figure 2 A representative example of a simple 3-fold complex for different and separate components is shown;

[0015] Figure 3A A two-dimensional representative example of an entity embedding set is shown;

[0016] Figure 3B A representative example of a defined ε (sphere radius) according to one embodiment is shown;

[0017] Figure 3C The use according to one embodiment is shown. Figures 3A-3B The examples in the text are representative examples of simple complex constructions;

[0018] Figure 4 A block diagram illustrating the flow of a machine learning model architecture for cross-domain structured mapping according to one embodiment is shown;

[0019] Figure 5 A block diagram is shown of a process for cross-domain structured mapping to relate two or more corpora of dissimilar data, according to one embodiment;

[0020] Figure 6 A cloud computing environment according to an embodiment is described;

[0021] Figure 7 A set of abstract model layers according to the embodiments is described;

[0022] Figure 8 This is the network architecture of a system for cross-domain structured mapping according to an embodiment;

[0023] Figure 9 The embodiments shown can be used with Figure 6 The representative hardware environment associated with the server and / or client; and

[0024] Figure 10 This is a block diagram illustrating a distributed system for cross-domain structured mapping according to one embodiment. Detailed Implementation

[0025] Various embodiments have been described for illustrative purposes, but such description is not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein has been chosen to best explain the principles, practical application, or technical improvements of the embodiments relative to technologies found on the market, or to enable those skilled in the art to understand the embodiments disclosed herein.

[0026] The embodiments relate to cross-domain structured mapping machine learning models and systems. One embodiment provides a method using a computational device to correlate two or more corpora of dissimilar data, the method comprising receiving input data from each of the two or more corpora of dissimilar data. The computational device computes a transfer of each of the input data into two or more encoder-decoder models. The computational device also obtains a prediction of identity mapping for each distinct knowledge domain from each of the two or more encoder-decoder models. The computational device additionally computes a distributional distance metric as the output of each of the low-dimensional embedding vector representations from each of the two or more encoder-decoder models. The computational device also computes a function based on the predictions from each of the two or more encoder-decoder models and each of the distributional distance metrics. The computational device further updates the two or more encoder-decoder models. One or more of the following features may be included. In some embodiments, the method may further include having the computational device compute a corresponding reconstruction loss for each of the two or more encoder-decoder models using the corresponding predictions and input data from each of the two or more corpora of dissimilar data. The computing device may also include a low-dimensional embedding vector that extracts the input data representation from each of two or more encoder-decoder models.

[0027] In some embodiments, the method may further include: the distribution distance metric is a pairwise average relative survival time (MRLT) distribution distance metric, and the function is a joint loss function.

[0028] In one or more embodiments, the method may further include having a computing device compute the gradient of the loss from the joint loss function for the model parameters of each of two or more encoder-decoder models.

[0029] In some embodiments, the method may further include: initializing the weights of each encoder-decoder model in the two or more encoder-decoder models by the computing device. The computing device further performs preprocessing, transformation, and extraction of the input data into fixed-dimensional feature vectors. The computing device further performs feedforward processing to feedforward each in-domain sample of the input data to each corresponding model in the two or more encoder-decoder models. The computing device further uses each corresponding encoder-decoder model in the two or more encoder-decoder models to generate a corresponding output prediction for each in-domain sample of the input data. Given the in-domain samples of the input data and the corresponding output prediction, the computing device also computes a corresponding loss value with respect to the joint loss function of each of the two or more encoder-decoder models. In one or more embodiments, the method may include having the computing device compute a pairwise MRLT distribution distance metric based on a first relative survival time (RLT) matrix and a second RLT matrix between each in-domain sample of the input data and based on the distribution distance between the first RLT matrix and the second RLT matrix. In some embodiments, the method may further include computing a pairwise MRLT distribution distance metric based on a first RLT matrix and a second RLT matrix between samples within each domain of the input data and based on a squared loss function between the outputs of the first RLT matrix and the outputs of the second RLT matrix. In one or more embodiments, the method may include computing a pairwise MRLT distribution distance metric based on a computing device based on an RLT matrix and a second RLT matrix between samples within each domain of the input data and based on a Wasserstein distance determination using the distributions of the first RLT matrix and the second RLT matrix. In some embodiments, the method may include two or more corpora of the dissimilar data comprising text, images, audio, and other data sources in different knowledge domains.

[0030] One or more embodiments include an autoencoder employing one or more artificial intelligence (AI) models (e.g., autoencoder model 1 920 and autoencoder model 2 925). Figure 4 The model (e.g., model architecture 900) Figure 4 AI models can include trained ML models (e.g., neural networks (NN), convolutional NN (CNN), recurrent NN (RNN), long short-term memory (LSTM) based NN, gated recurrent unit (GRU) based RNN, tree-based CNN, self-attention networks (e.g., NNs that utilize attention mechanisms as basic building blocks; self-attention networks have shown effectiveness for sequence modeling tasks without recursion or convolution), BiLSTM (bidirectional LSTM), etc.). Artificial NNs are interconnected groups of nodes or neurons.

[0031] The process of cross-domain mapping originates from the application of lateral thinking, a methodology for solving problems using indirect and creative approaches that guide the thinking process through non-obvious analogies. In some embodiments, examples of lateral thinking applied in innovation include restaurants using conveyor belts to serve food choices, which can be based on the core concept of airport baggage carousels using conveyor belts to transfer objects. Another example of cross-domain innovation is a game controller as a user interface, which can be based on automotive innovation using a controller on a display to control car features via the concept of an intuitive controller. Both examples illustrate the principles of similar concepts, but in different contexts.

[0032] In some embodiments, cross-domain structured mapping can address the following example of use cases. In one example, given journal databases from two orthogonal domains (e.g., neuroscience and AI), suppose someone wants to find which subsets of papers are similar to each other. By identifying subset pairs of papers, in one embodiment, cross-domain structured mapping can help researchers identify and enhance their literature review process by discovering new combinations of ideas between the two domains. The methods of the embodiments significantly improve the efficiency of researchers in simplifying cross-disciplinary source discovery and matching without requiring significant knowledge of other external domains. As a result, the embodiments significantly improve new discoveries and connections between two or more different fields that may not be easily identified through human work and can be performed rapidly by one or more embodiments.

[0033] During the bidding process, clients submit Request for Proposal (RFP) documents. In each phase, multiple documents exist describing client needs, the bidding process, logistics, etc. Competitors / bidders, acting as service or product providers, need to consume resources to understand and extract relevant client requirements and bidding information. These competitors / bidders then draft proposals for saleable items that meet the client requirements based on the information they have extracted. Typically, this process is manual; it requires expert knowledge and is error-prone and labor-intensive. Several state-of-the-art (SOTA) methods exist for the cognitive extraction of requirements from RFP documents. However, SOTA methods do not jointly extract and inject requirements from multiple sources into a single response proposal. Some embodiments provide different components for jointly extracting and associating RFP documents and mapping them to the processing of relevant saleable items and other products / services (e.g., timelines, delivery methods, etc.) that will be included in the RFP document.

[0034] In another example, given a set of business problems or processes (e.g., a Component Business Model (CBM) portal) and a list of codebases and research assets (e.g., existing for-sale or research assets of an entity), some embodiments facilitate this by mapping requirements from clients to existing solutions offered by developers and researchers. The approach of these embodiments significantly improves the discovery of new assets (including new use cases for certain solutions) from the client side. Another benefit is that researchers can discover new applications of their work in new fields and are encouraged to reuse their work and assets in different areas, thus promoting efficient utilization.

[0035] Figure 1 Example 600 is shown, according to one embodiment, which uses topological representations to compare different representations of a domain, employing a defined structure to jointly learn appropriate mappings between these representations. In some embodiments, key components for learning and comparing inter-domain structural similarity between different domains require learning robust representations with the following properties: the ability to estimate the quality and diversity of the representation for each domain; modeling the complex and nonlinear structures (relationships between entities) within each entity represented in the domain; and providing a method for comparing between two independent domains of an entity set. Example 600 includes a first domain (domain 1 610) P. D1 (X1) and the second domain (domain 2 650) P D2 (X2). Domain 1 610 and Domain 2 650 each contain multiple documents.

[0036] A manifold is a topological space analogous to the Euclidean space around each point. A topological space can be defined as a set of points and a set of neighborhoods for each point, satisfying a set of axioms associated with the points and their neighborhoods. Each point in an n-dimensional manifold has a neighborhood homeomorphic to a Euclidean space of dimension n. The dimension of a mathematical space (or object) can be defined as the minimum number of coordinates required for any point within a specified dimension. Some embodiments utilize a focus on low-dimensional manifolds M. data Data p on data The distribution of (x) is assumed to be based on the following assumptions: the manifold is a complex nonlinear structure; and the data p data New features and patterns of (x) can be found in the properties of manifolds (such as loops and M). data This involves representing higher-dimensional holes. One embodiment utilizes a topological representation that provides a way to compare different representations of a domain using a defined structure and jointly learn appropriate mappings between these representations. In Example 600, domain 1 610 is represented as manifold 1 620 (M1) and domain 2 650 is represented as manifold 2 640 (M2). In one embodiment, a cross-domain structured mapping process (described below) minimizes the topological distance 630 between manifold 1 620 and manifold 2 640.

[0037] Figure 2 A representative example of a simple 3-complex 700 for different and separate components is shown. Because there is no direct access to the topological representation of the data, one embodiment provides an approximate representation of the manifold learned based on data samples present for each domain. One example embodiment uses a simpler representation space, such as a simple complex, which is a representation composed of points, line segments, triangles, and higher-order tetrahedrons. An n-dimensional simplex is the convex hull of n+1 affine points. In the example simple 3-complex 700, it can be seen that the maximum dimension of the simple complex is three-dimensional (4 faces) because the tetrahedron has four corners. Therefore, in this example, it is considered a simple "3"-complex.

[0038] Figure 3A A representative example of entity embedding set 810 is shown. Using the representation in this example, cross-domain structured mapping processing considers the relevant connections between pairwise distances between individual samples.

[0039] Figure 3B A two-dimensional representative example 820 of a defined ε (sphere radius) 825 according to one embodiment is shown. When defining ε (sphere radius), it is assumed that an ε (sphere radius) 825 provides a single simplicial complex, and one embodiment considers different ranges of ε values. In one embodiment, a family of simplicial complexes is considered, and for each of the different ε values, a cross-domain structured mapping process quantifies the properties of the simplicial complex by evaluating emergent features or homology (such as the number of loops and holes). For each different ε value, there exists a different number of homology values ​​that control which relationships are important and which are noisy. In one embodiment, these formed k homology values ​​(e.g., how many k-dimensional holes exist) are appropriately ordered by observing the intersection of each bar in a persistent barcode (a graphical representation in a bar graph format representation of simplicial components) (where each bar in the persistent barcode represents a component in a simplicial complex; or a grouping of data).

[0040] Figure 3C The use according to one embodiment is shown. Figures 3A-3B The example in the example is a representative example of the simplicial complex construction 830. In one embodiment, the ε parameter is utilized as a process for determining the optimal sphere radius, which provides a mapping for determining the optimal simplicial complex (a representation of intra-domain relations) to perform such relations across domains. In some embodiments, there is a need to efficiently determine persistent barcodes for large datasets and compare them between different persistent barcodes (which is non-trivial in many cases). In one embodiment, a small subset of landmarks is used to construct the simplicial complex while considering proximity points known as witnesses. The witness complex is... Simplexes are defined on two sets of points in a set of real numbers in D dimensions. Landmarks are the vertices of the simplexes, and witnesses help determine which simplexes are "witnessed" (observed).

[0041] One problem with conventional methodologies is how to learn representations, which is unclear. Specifically, the first step of taking pairwise distances to learn relationships between entities is not scalable and is not well executed, which is conventionally suggested to include an "end-to-end" approach. One conventional approach proposes using the topological properties of the low-dimensional embedding representation of the data as a comparison between two probability distributions, comparing the quality and diversity of the generative adversarial network (GAN) as generated data against real data. In one embodiment, the cross-domain structured mapping process learns a (good) topological representation (which can be represented as a probability distribution); attempts to perform a cross-domain structured mapping on the topological space (rather than the geometric space); and the persistent barcode representation is used to indicate the use of an RLT, which is used as a metric to indicate the number of holes, or equivalently, the number of connected components found in the dataset relative to the radius of a sphere of a simplex. One or more embodiments contribute to the advantage of using topological features as part of the cross-domain mapping loss function. In one embodiment, the geometric score is used as part of the joint loss function, rather than the auxiliary scoring function used in conventional techniques. Another advantage of one or more embodiments is that the process requires far fewer samples than conventional techniques, making the cross-domain structured mapping process tractable for large datasets.

[0042] Figure 4 A block diagram illustrating the flow of an ML model architecture 900 for cross-domain structured mapping processing according to one embodiment is shown. In one embodiment, the ML model architecture 900 includes a domain corpus (D1) 905, a domain corpus (D2) 906, and input data 915. Input data 916 Autoencoder model 1920 (with encoder 921 and decoder 922), autoencoder model 2925 (with encoder 926 and decoder 927), output prediction 930 Output prediction 931 The architecture includes reconstruction loss processing (935), reconstruction loss processing (936), loss function processing (940), distribution distance metric processing (945), RLT processing (950), domain 1 embedding vector processing (960), and domain 2 embedding vector processing (965). Typically, this architecture can be scaled to N corpora, thereby scaling the number of cross-domain mappings to perform across N different corpora and N autoencoder models.

[0043] In one embodiment, the cross-domain structured mapping processing of the ML model architecture uses the RLT of each hole observed in the simplex complex to extract an estimate of the cohomology. RLT is the ratio of the total time the cohomology exists (where ε is considered the time axis) to the maximum ε (the point where the complex becomes a single entity). RLT can also assess how well the manifold representation is approximated with confidence. To ensure robustness, in one embodiment, RLT is modeled randomly by considering the mean or average RLT (MRLT) – by selecting some landmarks or landmark points (based on a scalable complex suitable for a large set of points, e.g., a witness complex). In this way, a probability distribution for each cohomology is derived, which can then allow comparisons across other manifold representations; through measures such as L2 error or Wasserstein line distance. In one embodiment, RLT is used as part of the (parameterized) loss function processing 940, rather than as an auxiliary metric used in conventional techniques.

[0044] In one embodiment, RLT processing 950 can be represented as follows:

[0045]

[0046] Where i is the persistence interval, k is the Kth dimension of the simplicium, X is the dataset, L is the landmark, μ is the mean calculation, and α∈[0, α...]. max ] is the range of ε (α) max (This is the upper limit of ε), β k (α) = i is the Kth Betti number. Given a value of α, which provides a simplicial complex with a specific homology, the Betti number gives an integer count of the persistence interval containing α:

[0047]

[0048] Where b and d represent the start and end points of the persistence interval, respectively. Let i represent the set of n persistent intervals in the k-th dimension, where i and k are integers.

[0049] In one embodiment, MRLT is represented as follows:

[0050]

[0051] in It is the expectation of the RLT at a given landmark L. This leads to the conclusion that the MLRT can be defined by the following probability distribution:

[0052]

[0053] In one embodiment, the input data (input data 915) Input data 916 (etc.) may include a set of N separate domain entity sets (Di) containing a certain number of items (e.g., documents, nodes in a graph, images, videos, code, etc.). In one embodiment, the output (output prediction 930) Output prediction 931 (etc.) includes a transformation function or mapping model learned between intra-domain entity vectors across N domains. In other embodiments, the output mapping model may be further used as an index or transformation function for a search engine. ML model architecture 900 provides a learned model that maps between intra-domain structural differences of entities within each domain corpus, such that the resulting embedding model can generate an index for any arbitrary cross-domain input.

[0054] In one embodiment, autoencoder models 1 920 and 2 925 may include four parts: an encoder (e.g., encoders 921 and 926), where the autoencoder model learns how to reduce the input dimensionality and compress the input data into an encoded representation; a bottleneck, which is a layer containing the compressed representation of the input data (the lowest possible dimensionality of the input data); a decoder (e.g., decoder 922, decoder 927), where the model learns how to reconstruct data from the encoded representation as closely as possible to the original input; and a reconstruction loss (reconstruction loss processing 935, reconstruction loss 936), which is a process that measures how well the decoder performs and how close the output is to the original input. Training autoencoder models 1 920 and 2 925 involves using backpropagation to minimize the reconstruction loss.

[0055] In one embodiment, the autoencoder models (autoencoder model 1 920 and autoencoder model 2 925) may each include an RNN encoder-decoder pair, which comprises two RNNs acting as an encoder and decoder pair. The encoder (e.g., encoder 921, encoder 926) maps variable-length source sequences to fixed-length vectors, and the decoder (e.g., decoder 922, decoder 927) maps the vector representation back to the variable-length target sequence. In some embodiments, the autoencoder models are all unsupervised artificial neural networks that learn how to efficiently compress and encode data, and then learn how to reconstruct the data from the simplified encoded representation back to a representation as close as possible to the original input.

[0056] In one embodiment, the processing of the ML model architecture 900 involves computing the feedforward of the autoencoder models (autoencoder models 1920 and 2925), computing the RLT of the latent embeddings (RLT processing 950), computing the RLT distribution distance loss metric, and computing the joint loss function (loss function processing 940), gradients, and updating the model, as described below.

[0057] In one embodiment, calculating the feedforward of the autoencoder models (autoencoder model 1920 and autoencoder model 2925) includes the following. In one embodiment, the input data (input data 915) Input data 916 (etc.) including domain entities (D i The set of N separate sets contains a certain number of items (e.g., documents, nodes in a graph, images, videos, code, etc.). The output of this processing part includes the predicted output of the corresponding (in-domain) autoencoder model (autoencoder model 1920 and autoencoder model 2925) from each item in set D, and the corresponding loss value for each prediction.

[0058] In one embodiment, for feedforward propagation, given N individual domain entities (D i The processing first feeds each sample into its corresponding autoencoder model (autoencoder model 1920 and autoencoder model 2925) to generate a prediction and compute its loss value accordingly. In some embodiments, the autoencoder models (autoencoder model 1920 and autoencoder model 2925) may be shallow autoencoders or seq-to-seq encoder-decoder autoencoder models. For a given prediction, the ML model architecture 900 also computes the corresponding loss value for each defined model.

[0059] In one embodiment, for the feedforward pass, for each domain's autoencoder model (autoencoder model 1920 and autoencoder model 2925), weights are initialized for the corresponding autoencoder model. Next, for the input data (input data 915)... Input data 916 The data is preprocessed and transformed, and then extracted into a fixed-dimensional feature vector. Next, for each in-domain sample, the feedforward processing is executed into the corresponding autoencoder model (note that the specific implementation depends on the embodiment of the ML model used). The corresponding in-domain autoencoder model (autoencoder model 1920 and autoencoder model 2925) generates a corresponding output prediction for each sample. Given the input sample data and the predicted output, the corresponding loss value is calculated with respect to a given loss function for the given model.

[0060] In one embodiment, the computation of the latent embedding RLT (RLT processing 950) computes an RLT for the intra-domain set of each embedding vector, which is later used to compute the distance between two RLTs for the intra-domain set of embedding vectors. In one embodiment, the computation of the latent embedding RLT includes the following as input: the set X representing the embedding vectors of each item in set D; the number L0 of landmarks to be used; and the value ε (α) used to determine the upper bound. max The coefficient γ; the parameter i used to determine the upper limit of the persistence interval. max ; iteration number n; distance function dist(a, b) for calculating the distance between sample a and sample b; witness(d, α, k) for calculating the family witness of the complex with pairwise distance d, the maximum persistence α, and the maximum dimension k of the simplex; and persistence(w, k) for calculating the persistence interval of the family with dimension k. In one embodiment, the computation of the latent embedding RLT includes the following as output: n times i including RLT measurements max matrix.

[0061] In one embodiment, the computational initialization of the latent embedded RLT has a dimension n multiplied by i. max The matrix is ​​used to store the computational results of RLT. Then, the process randomly selects L0 from the embedding vector representation of X and assigns it to L. Next, given L and X, the process computes a given distance function using the defined distance metric dist(L, X) and assigns it to d. The maximum ε-size is computed from γ·max(dist(L, L)) and assigned to α. max Given d and α max Next, we use the witness function witness(d, α) max 2) Determine the witness for the complex, where k = 2, and assign the output to W. Given W, determine the persistence value by computing persistence(W, 1), which is assigned to I. Using the value calculated above, determine the RLT (using RLT(i, X, L, K)), and for each sample and in [0, i... max The ε range between ] is filled into the RLT matrix.

[0062] In one embodiment, to compute the RLT distribution distance loss metric, the input includes the distribution of the RLT matrix for each domain from different data sources, and the output is a quantified measurement of the distance between the RLT in the first domain and the RLT in the second domain. In this part of the processing of the ML model architecture 900, the distance between the two distributions of the MRLT from each domain dataset is determined and then used as part of the loss function processing 940 of the ML model architecture 900 to minimize it.

[0063] In one embodiment, calculating the RLT distribution distance loss metric includes the following: given a first RLT matrix RLT(X) (1) ) and the second RLT matrix RLT(X (2) Then, the distance metric between the two domain-internal embedding datasets can be calculated. In one embodiment, the Wasserstein distance is used, where the processing considers calculating RLT(X). (1) ) and RLT(X (2) The distance between the distributions of ) . In an alternative embodiment, RLT(X) is implemented. (1) The output of ) and RLT(X (2) The squared loss function between the outputs of ).

[0064] In one embodiment, to compute the joint loss function (loss function processing 940), gradients, and update the ML model architecture 900, the inputs include the reconstruction loss function from each autoencoder model (autoencoder model 1920 and autoencoder model 2925) and the distance function from the RLT of each in-domain embedding vector. The outputs include the joint loss function from the ML model architecture 900 (from each of the different components) and the updated ML model architecture 900 parameters from the backpropagation processing. In this part of the processing of the ML model architecture 900, the joint loss function processing 940 is determined from the computation of all previously computed components, and then a stochastic gradient descent (SGD) process is performed to update the ML model architecture 900 parameters. This process jointly optimizes the reconstruction loss and the loss from the distance from the RLT metric. The SGD process includes an iterative process for optimizing the objective function by utilizing appropriate smoothness properties.

[0065] In one embodiment, to compute the joint loss function (loss function processing 940), gradients, and update the ML model architecture 900, given each loss function of the reconstruction loss (reconstruction loss processing 935, reconstruction loss processing 936) from the autoencoder models (autoencoder models 1920 and 2925) and the distance function between the in-domain embedding vectors RLT, each component is summed into a single loss value. Next, this process computes the gradients and performs SGD processing to update the parameters of the ML model architecture 900 relative to the original encoder-decoder models (autoencoder models 1920 and 2925).

[0066] One or more embodiments can apply the processing of the ML model architecture 900 as an enabling technology to further accelerate and enhance research and innovation workflows across a wide range of industries and technologies. Embodiments can use deep learning methods to enhance and bridge the gap between classical symbolic methods of AI—introducing techniques for advancing state-of-the-art (SOTA) solutions in hybrid approaches.

[0067] Figure 5 A block diagram of a process 1000 for cross-domain structured mapping to relate two or more corpora of dissimilar data, according to one embodiment, is shown. In one embodiment, in block 1010, process 1000 is performed by a computing device (from computing node 10, ... Figure 6 Hardware and software layers 60 Figure 7 Processing system 300 Figure 8 System 400 Figure 9 System 500 Figure 10 ML model architecture 900 Figure 4 (etc.) receive two or more corpora containing dissimilar data (e.g., domain corpus (D1) 905, domain corpus (D2) 906, etc.). Figure 4 Access to ) is provided. In box 1020, process 1000 further provides the encoding of each of two or more corpora of dissimilar data by a computing device (e.g., using encoder 921 of autoencoder model 1920, encoder 926 of autoencoder model 2925, etc.). Figure 4 In box 1030, process 1000 further provides the similarity of entities in each of two or more corpora of dissimilar data calculated by a computing device (e.g., using...). Figure 4 (RLT processing 950). In box 1040, process 1000 further provides the generation of mutually related entity corpora by a computing device based on the calculated similarity of entities in each of two or more corpora of dissimilar data.

[0068] In one embodiment, process 1000 may further include the feature that the two or more corpora of dissimilar data are scientific journal articles in different knowledge domains.

[0069] In one embodiment, process 1000 may additionally include receiving features of input data from each of two or more corpora of dissimilar data. The computing device further computes features of each input data into two or more encoder-decoder models (e.g., Figure 4 The computing device further obtains predictions of identity mappings for each of the two or more encoder-decoder models in different knowledge domains. The computing device also uses the corresponding predictions (e.g., output prediction 930) in the autoencoder models 1920 and 2925. Output prediction 931 The corresponding reconstruction loss for each of two or more encoder-decoder models is computed using input data from each of two or more dissimilar corpora (e.g., reconstruction loss processing 935 and reconstruction loss processing 936). Figure 4).

[0070] In one embodiment, process 1000 may further include extracting low-dimensional embedding vectors representing the input data from each of two or more encoder-decoder models by a computing device (e.g., from...). Figure 4 The computing device further computes the pairwise MRLT distribution distance metric as the output of each of the low-dimensional embedding vector representations from each of the two or more encoder-decoder models. The computing device also further computes a joint loss function (e.g., via loss function processing 940) based on each of the predictions from each of the two or more encoder-decoder models and the pairwise MRLT distribution distance metric. The computing device further computes the gradient of the loss from the joint loss function with respect to the model parameters of each of the two or more encoder-decoder models. The computing device further updates the two or more encoder-decoder models.

[0071] In one embodiment, process 1000 may further include initializing the weights of each of two or more encoder-decoder models by a computing device. The computing device further performs preprocessing, transformation, and extraction of the input data into fixed-dimensional feature vectors. The computing device also further performs feedforward processing to feedforward each in-domain sample of the input data to each corresponding model of the two or more encoder-decoder models. The computing device further uses each corresponding model of the two or more encoder-decoder models to generate a corresponding output prediction (e.g., output prediction 930) for each in-domain sample of the input data. Output prediction 931 Given in-domain samples of the input data and the corresponding output predictions, the computing device also computes the corresponding loss value for the joint loss function of each of the two or more encoder-decoder models.

[0072] In one embodiment, process 1000 may further include a computing device calculating a feature of a pairwise MRLT distribution distance metric based on a first RLT matrix and a second RLT matrix between samples within each domain of the input data and based on the distance between the distributions using the first RLT matrix and the second RLT matrix.

[0073] In one embodiment, process 1000 may further include a computing device calculating features of a pairwise MRLT distribution distance metric based on a first RLT matrix and a second RLT matrix between samples within each domain of the input data and based on a squared loss function using the outputs of the first RLT matrix and the second RLT matrix.

[0074] It should be understood beforehand that while this disclosure includes a detailed description of cloud computing, the implementation of the teachings cited herein is not limited to cloud computing environments. Rather, embodiments of this example can be implemented in conjunction with any other type of computing environment now known or developed hereafter.

[0075] Cloud computing is a service delivery model for enabling convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, network bandwidth, servers, processing, storage, applications, virtual machines (VMs), and services) that can be rapidly configured and released with minimal management effort or interaction with service providers. This cloud model may include at least five features, at least three service models, and at least four deployment models.

[0076] The features are as follows:

[0077] On-demand self-service: Cloud consumers can unilaterally and automatically provision computing power, such as server time and network storage, on demand without human interaction with service providers.

[0078] Extensive network access: Capabilities are available on the network and accessed through standard mechanisms that facilitate use by heterogeneous, thin or thick client platforms (e.g., mobile phones, laptops, and PDAs).

[0079] Resource pooling: Pooling a provider's computing resources to serve multiple consumers using a multi-tenant model, where different physical and virtual resources are dynamically allocated and reallocated as needed. Location independence is significant because consumers typically do not have control or knowledge of the exact location of the resources provided, but may be able to specify the location at a higher level of abstraction (e.g., country, state, or data center).

[0080] Rapid elasticity: Capacity can be provided quickly and flexibly, and in some cases, it can be automatically, rapidly scaled up, and rapidly released for rapid expansion. For consumers, the capacity available for supply often appears unlimited and can be purchased in any quantity at any time.

[0081] Measuring services: Cloud systems automatically control and optimize resource usage by leveraging metering capabilities at a level of abstraction appropriate to the service type (e.g., storage, processing, bandwidth, and consumer accounts). Resource usage can be monitored, controlled, and reported, providing transparency for both the providers and consumers of the services being utilized.

[0082] The business model is as follows:

[0083] Software as a Service (SaaS): The capability offered to consumers is the ability to use the provider's applications running on cloud infrastructure. Applications can be accessed from different client devices via thin client interfaces such as web browsers (e.g., web-based email). Consumers do not manage or control the underlying cloud infrastructure, including the network, servers, operating system, storage, or even individual application capabilities, with the possible exception of limited consumer-specific application configuration settings.

[0084] Platform as a Service (PaaS): This provides consumers with the ability to deploy applications created or acquired by the consumer onto cloud infrastructure using programming languages ​​and tools supported by the provider. Consumers do not manage or control the underlying cloud infrastructure, including networks, servers, operating systems, or storage, but they have control over the deployed applications and, possibly, the configuration of the application hosting environment.

[0085] Infrastructure as a Service (IaaS): This provides consumers with the ability to provide processing, storage, networking, and other basic computing resources in which consumers can deploy and run any software, including operating systems and applications. Consumers do not manage or control the underlying cloud infrastructure, but rather have control over the operating system, storage, deployed applications, and possibly limited control over selected networking components (e.g., host firewalls).

[0086] The deployment model is as follows:

[0087] Private cloud: A cloud infrastructure for organization operations only. It can be managed by the organization or a third party and can exist on-site or off-site.

[0088] Community cloud: A cloud infrastructure shared by several organizations and supporting a specific community with shared concerns (e.g., tasks, security requirements, policies, and compliance considerations). It can be managed by an organization or a third party and can exist on-site or off-site.

[0089] Public cloud: Cloud infrastructure that is made available to the public or large groups of industries and is owned by an organization that sells cloud services.

[0090] Hybrid cloud: A cloud infrastructure is composed of two or more clouds (private, community, or public) that remain a single entity but are bound together by standardized or proprietary technologies that enable data and application portability (e.g., cloud bursting for load balancing between clouds).

[0091] Cloud computing environments are service-oriented and focus on statelessness, loose coupling, modularity, and semantic interoperability. The core of cloud computing is its infrastructure, which includes a network of interconnected nodes.

[0092] See now Figure 6The diagram illustrates an illustrative cloud computing environment 50. As shown, the cloud computing environment 50 includes one or more cloud computing nodes 10, whose local computing devices used by cloud consumers (such as personal digital assistants (PDAs) or mobile phones 54A, desktop computers 54B, laptop computers 54C, and / or automotive computer systems 54N) can communicate with the cloud computing nodes 10. The nodes 10 can communicate with each other. They can be physically or virtually grouped (not shown) in one or more networks, such as private clouds, community clouds, public clouds, or hybrid clouds, as described above, or combinations thereof. This allows the cloud computing environment 50 to provide infrastructure, platform, and / or software as services that cloud consumers do not need to maintain on their local computing devices. It should be understood that... Figure 6 The types of computing devices 54A-N shown are intended to be illustrative only, and computing node 10 and cloud computing environment 50 can communicate with any type of computerized device via any type of network and / or network-addressable connectivity (e.g., using a web browser).

[0093] See now Figure 7 This demonstrates a cloud computing environment of 50 ( Figure 6 This provides a set of functional abstractions. It should be understood beforehand that... Figure 7 The components, layers, and functions shown are intended to be illustrative only, and the embodiments are not limited thereto. As depicted, the following layers and corresponding functions are provided:

[0094] The hardware and software layer 60 includes hardware and software components. Examples of hardware components include: a host 61; a server 62 based on a RISC (Reduced Instruction Set Computer) architecture; a server 63; a blade server 64; a storage device 65; and network and networking components 66. In some embodiments, software components include network application server software 67 and database software 68.

[0095] The virtualization layer 70 provides an abstraction layer from which the following examples of virtual entities can be provided: virtual server 71; virtual storage 72; virtual network 73, including virtual private network; virtual application and operating system 74; and virtual client 75.

[0096] In one example, management layer 80 provides the functionality described below. Resource provisioning 81 provides dynamic acquisition of computing resources and other resources for performing tasks within the cloud computing environment. Metering and pricing 82 provides cost tracking as resources are utilized within the cloud computing environment and bills or invoices for the consumption of these resources. In one example, these resources may include application software licenses. Security provides authentication for cloud consumers and tasks, as well as protection for data and other resources. User portal 83 provides access to the cloud computing environment for consumers and system administrators. Service level management 84 provides cloud resource allocation and management to ensure the required service level is met. Service level agreement (SLA) planning and fulfillment 85 provides pre-scheduling and procurement of cloud resources, anticipating future demand for those resources according to the SLA.

[0097] Workload tier 90 provides examples of functionalities that can leverage a cloud computing environment. Examples of workloads and functionalities that can be provided from this tier include: mapping and navigation 91; software development and lifecycle management 92; virtual classroom education delivery 93; data analytics and processing 94; transaction processing 95; and cross-domain structured mapping processing 96 (see, for example, System 500). Figure 10 ML model architecture 900 for cross-domain structured mapping Figure 4 And processing 1000, Figure 5 As mentioned above, regarding Figure 7 All the foregoing examples described are illustrative only, and the embodiments are not limited to these examples.

[0098] To reiterate, although this disclosure includes a detailed description of cloud computing, implementations of the teachings cited herein are not limited to cloud computing environments. Rather, the embodiments can be implemented using any type of cluster computing environment now known or developed hereafter.

[0099] Figure 8 This is the network architecture of system 300 for cross-domain structured mapping according to an embodiment. For example... Figure 8 As shown, multiple remote networks 302 are provided, including a first remote network 304 and a second remote network 306. A gateway 301 can be coupled between the remote network 302 and a neighboring network 308. In the context of this network architecture 300, networks 304 and 306 can each take any form, including but not limited to LAN, WAN (such as the Internet), Public Switched Telephone Network (PSTN), internal telephone network, etc.

[0100] In use, gateway 301 serves as the entry point from remote network 302 to neighboring network 308. Accordingly, gateway 301 can function as both a router and a switch; a router directs a given data packet to gateway 301, while a switch provides the actual path for a given packet to and from gateway 301.

[0101] It also includes at least one data server 314 coupled to a neighboring network 308, which is accessible from a remote network 302 via a gateway 301. It should be noted that the data server 314 may include any type of computing device / groupware. Coupled to each data server 314 are multiple user devices 316. Such user devices 316 may include desktop computers, laptop computers, handheld computers, printers, and / or any other type of device containing logic. It should be noted that in some embodiments, the user devices 316 may also be directly coupled to any one of the devices in the network.

[0102] Peripheral 320 or a series of peripherals 320 (e.g., fax machine, printer, scanner, hard drive, networking and / or local storage unit or system, etc.) may be coupled to one or more of networks 304, 306, and 308. It should be noted that databases and / or additional components may be used with or integrated into any type of network element coupled to networks 304, 306, and 308. In the context of this specification, "network element" may refer to any component of the network.

[0103] According to some methods, the methods and systems described herein can be implemented together with and / or on a virtual system and / or a system that simulates one or more other systems, such as simulation... z / OS environment System, virtual hosting Environment System, simulation Environment Systems, etc. In some embodiments, this can be achieved by using... Software is used to achieve this virtualization and / or simulation.

[0104] Figure 9 An illustration of a method according to one embodiment is shown. Figure 8 The representative hardware system 400 environment associated with user equipment 316 and / or server 314. In one example, the hardware configuration includes a workstation having a central processing unit 410 such as a microprocessor and several other units interconnected via a system bus 412. Figure 9The workstation shown may include random access memory (RAM) 414, read-only memory (ROM) 416, I / O adapter 418 for connecting peripheral devices (e.g., disk storage unit 420) to bus 412, user interface adapter 422 for connecting keyboard 424, mouse 426, speaker 428, microphone 432 and / or other user interface devices, such as touch screen, digital camera (not shown), communication adapter 434 for connecting the workstation to communication network 435 (e.g., data processing network), and display adapter 436 for connecting bus 412 to display device 438.

[0105] In one example, the workstation may have an operating system residing on it, such as Operating System (OS), MAC OS, etc. In one embodiment, system 400 adopts a system based on... The file system. It will be understood that other examples can also be implemented on platforms and operating systems other than those mentioned. Such other examples may include using... Operating systems written in XML, C and / or C++ or other programming languages, and object-oriented programming methods. Object-oriented programming (OOP) can also be used, and it is increasingly being used to develop complex applications.

[0106] Figure 10 This is a block diagram illustrating a distributed system 500 for cross-domain structured mapping according to one embodiment. In one embodiment, system 500 includes client devices 510 (e.g., mobile devices, smart devices, computing systems, etc.), a cloud or resource-sharing environment 520 (e.g., public cloud computing environment, private cloud computing environment, data center, etc.) and a server 530. In one embodiment, client devices 510 are provided with cloud services from server 530 through the cloud or resource-sharing environment 520.

[0107] One or more embodiments may be systems, methods, and / or computer program products at any possible level of integration technical detail. A computer program product may include a computer-readable storage medium having computer-readable program instructions thereon for causing a processor to execute aspects of this embodiment.

[0108] Computer-readable storage media can be tangible devices capable of retaining and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital universal disc (DVD), memory sticks, floppy disks, mechanical encoding devices such as punch cards or having raised structures in grooves on which instructions are recorded, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0109] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a corresponding computing / processing device, or downloaded to an external computer or external storage device via a network (e.g., the Internet, a local area network, a wide area network, and / or a wireless network). The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the corresponding computing / processing device.

[0110] Computer-readable program instructions used to perform the operations of the embodiments may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​(such as Smalltalk, C++, etc.) and procedural programming languages ​​(such as the "C" programming language or similar programming languages). The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter case, the remote computer may be connected to the user's computer via any type of network (including a local area network (LAN) or a wide area network (WAN)) or may be connected to an external computer (e.g., via the Internet through an Internet service provider). In some embodiments, electronic circuitry (including, for example, programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs)) may execute the computer-readable program instructions by utilizing state information from the computer-readable program instructions to personalize the electronic circuitry in order to perform aspects of this embodiment.

[0111] This document describes aspects of embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products. It should be understood that each block of the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer-readable program instructions.

[0112] These computer-readable program instructions may be provided to a processor of a computer or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in one or more blocks of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can instruct a computer, programmable data processing apparatus, and / or other device to operate in a particular manner, such that the computer-readable storage medium having the instructions stored therein includes an article of writing comprising instructions for implementing aspects of the functions / actions specified in one or more blocks of a flowchart and / or block diagram.

[0113] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus or other device, perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0114] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments. Each block in a flowchart or block diagram may represent a portion of a module, segment, or instruction, including one or more executable instructions for implementing a specified logical function. In some alternative embodiments, the functions marked in the blocks may occur in a different order than indicated in the figures. For example, depending on the functions involved, two consecutively shown blocks may actually be performed as a single step, executed simultaneously, substantially simultaneously, or in a manner that overlaps partially or entirely in time, or these blocks may sometimes be executed in reverse order. It will also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

[0115] Unless expressly stated otherwise, references to elements in the singular form in the claims are not intended to mean "one and only one," but rather "one or more." All structural and functional equivalents of the elements of the exemplary embodiments described above, which are now known or will later become apparent to those skilled in the art, are intended to be covered by the claims. Elements of the claims herein should not be construed under paragraph 6 of Section 112 of 35 U.S.SC unless the element is explicitly stated using the phrases "means for..." or "steps for...".

[0116] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the embodiments. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that when the terms “comprising” and / or “including” are used in this specification, they specify the presence of the stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or groups thereof.

[0117] All means or steps in the following claims, plus corresponding structures, materials, actions, and equivalents of the functional elements, are intended to include any structure, material, or action for performing the function in conjunction with other claimed elements as specifically claimed. The description of this embodiment has been presented for purposes of illustration and description, but is not intended to be exhaustive or limited to the embodiments of the disclosed forms. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the embodiments. The embodiments were chosen and described in order to best explain the principles and practical application of the embodiments, and to enable those skilled in the art to understand embodiments with different modifications suitable for the intended particular use.

Claims

1. A method of using a computing device to correlate two or more domain corpora of dissimilar data, the method comprising: Receive input data from each of two or more domain corpora containing dissimilar data; The computing device calculates the transmission of each of the input data; Based on the transmission of each of the input data, train two or more encoder-decoder models; The computing device obtains predictions for identity mapping for each different knowledge domain from each of the two or more encoder-decoder models; The computing device computes a distribution distance metric as output from each low-dimensional embedding vector representation of each of the two or more encoder-decoder models; The computing device computes a joint loss function based on each of the predictions from each of the two or more encoder-decoder models and the distribution distance metric; The computing device updates the training of the two or more encoder-decoder models based on the result of the joint loss function, wherein the training includes using backpropagation to minimize the reconstruction loss; The ε parameter is used to determine the sphere radius to determine the simplex complex representation of intradomain relations in order to perform the mapping of interdomain relations; as well as Using the updated two or more encoder-decoder models and the sphere radius, a cross-domain structured mapping is determined that correlates dissimilar data of the two or more encoder-decoder models.

2. The method according to claim 1, further comprising: The computing device uses the input data from each of the two or more domain corpora, including corresponding predictions and data from dissimilar data, to compute the corresponding reconstruction loss for each of the two or more encoder-decoder models; and The computing device extracts a low-dimensional embedding vector representing the input data from each of the two or more encoder-decoder models.

3. The method of claim 2, wherein the distribution distance metric is a pairwise average relative survival time (MRLT) distribution distance metric, and the function is a joint loss function.

4. The method according to claim 3, further comprising: The computing device calculates the gradient of the loss from the joint loss function relative to the model parameters of each of the two or more encoder-decoder models.

5. The method according to claim 4, further comprising: The computing device initializes the weights of each of the two or more encoder-decoder models; The computing device performs preprocessing, transformation, and extraction of the input data into fixed-dimensional feature vectors. The computing device performs feedforward processing to feedforward samples of the input data in each domain to each of the two or more encoder-decoder models. The computing device uses each of the two or more encoder-decoder models to generate a corresponding output prediction for each of the in-domain samples of the input data; and Given the in-domain samples of the input data and the corresponding output prediction, the computing device calculates the corresponding loss value with respect to the joint loss function for each of the two or more encoder-decoder models.

6. The method according to claim 4, further comprising: The computing device calculates the pairwise average relative survival time (MRLT) distribution distance metric based on a first relative survival time (RLT) matrix and a second relative survival time (RLT) matrix between samples in each domain of the input data, and based on two relative survival time metrics defined between the first and second relative survival time (RLT) matrices.

7. The method according to claim 4, further comprising: The computing device calculates the pairwise average relative survival time (MRLT) distribution distance metric based on a first relative survival time (RLT) matrix and a second relative survival time (RLT) matrix between samples in each domain of the input data, and based on a squared loss function using the output of the first relative survival time (RLT) matrix and the output of the second relative survival time (RLT) matrix.

8. The method according to claim 4, further comprising: The pairwise average relative survival time (MRLT) distribution distance metric is calculated by the computing device based on the first relative survival time (RLT) matrix and the second relative survival time (RLT) matrix between samples in each domain of the input data and based on the Wasserstein distance of the distribution using the first relative survival time (RLT) matrix and the second relative survival time (RLT) matrix.

9. The method of claim 1, wherein the two or more domain corpora of dissimilar data comprise text, images, audio, and other data sources in different knowledge domains.

10. A computer program product for associating two or more domain corpora of dissimilar data, the computer program product comprising program instructions executable by a processor to cause the processor to: The processor receives input data from each of two or more domain corpora containing dissimilar data. The processor calculates the transmission of each of the input data; Based on the transmission of each of the input data, train two or more encoder-decoder models; The processor obtains predictions for identity mapping for each different knowledge domain from each of the two or more encoder-decoder models; The processor computes a distribution distance metric as output from each low-dimensional embedding vector representation of each of the two or more encoder-decoder models; The processor computes a joint loss function based on each of the predictions from each of the two or more encoder-decoder models and the distribution distance metric; The processor updates the training of the two or more encoder-decoder models based on the result of the joint loss function, wherein the training includes using backpropagation to minimize the reconstruction loss; The ε parameter is used to determine the sphere radius to determine the simplex complex representation of intradomain relations in order to perform the mapping of interdomain relations; as well as Using the updated two or more encoder-decoder models and the sphere radius, a cross-domain structured mapping is determined that correlates dissimilar data of the two or more encoder-decoder models.

11. The computer program product according to claim 10, wherein: The program instructions can be executed by the processor to further cause the processor to: The processor computes the corresponding reconstruction loss for each of the two or more encoder-decoder models using the corresponding predictions and the input data from each of the two or more domain corpora of dissimilar data; and The processor extracts a low-dimensional embedding vector representing the input data from each of the two or more encoder-decoder models; and The two or more domain corpora of dissimilar data include text, images, audio, and other data sources from different knowledge domains.

12. The computer program product of claim 11, wherein the program instructions are executable by the processor to further cause the processor to: The processor calculates the gradient of the loss from the joint loss function relative to the model parameters of each of the two or more encoder-decoder models; The distribution distance metric is the pairwise average relative survival time (MRLT) distribution distance metric, and the function is the joint loss function.

13. The computer program product of claim 12, wherein the program instructions are executable by the processor to further cause the processor to: The processor initializes the weights of each of the two or more encoder-decoder models; The processor preprocesses, transforms, and extracts the input data into a fixed-dimensional feature vector. The processor performs feedforward processing to feedforward samples of the input data in each domain to each of the two or more encoder-decoder models. The processor uses each of the two or more encoder-decoder models to generate a corresponding output prediction for each of the in-domain samples of the input data; and Given the in-domain samples of the input data and the corresponding output prediction, the processor computes the corresponding loss value with respect to the joint loss function for each of the two or more encoder-decoder models.

14. The computer program product of claim 12, wherein the program instructions are executable by the processor to further cause the processor to: The processor calculates the pairwise average relative survival time (MRLT) distribution distance metric based on the first and second relative survival time (RLT) matrices between samples in each domain of the input data and based on the distance between the distributions of these two relative survival time metrics defined between the first and second relative survival time matrices.

15. The computer program product of claim 12, wherein the program instructions are executable by the processor to further cause the processor to: The processor calculates the pairwise average relative survival time (MRLT) distribution distance metric based on a first relative survival time (RLT) matrix and a second relative survival time (RLT) matrix between samples in each domain of the input data, and based on a squared loss function between the outputs of the first and second relative survival time (RLT) matrices.

16. The computer program product of claim 12, wherein the program instructions are executable by the processor to further cause the processor to: The pairwise average relative survival time (MRLT) distribution distance metric is calculated by the processor based on the first relative survival time (RLT) matrix and the second relative survival time (RLT) matrix between samples in each domain of the input data and based on the Wasserstein distance of the distribution using the first relative survival time (RLT) matrix and the second relative survival time (RLT) matrix.

17. An apparatus for relating two or more domain corpora of dissimilar data, comprising: The memory is configured to store instructions; as well as The processor is configured to execute the instructions to: Receive input data from each of two or more domain corpora containing dissimilar data; The processor calculates the transmission of each of the input data; Based on the transmission of each of the input data, train two or more encoder-decoder models; From each of the two or more encoder-decoder models, a prediction for identity mapping for each different knowledge domain is obtained; Compute the distribution distance metric as output from each low-dimensional embedding vector representation from each of the two or more encoder-decoder models; The joint loss function is computed based on each of the predictions from each of the two or more encoder-decoder models and the distribution distance metric; The training of the two or more encoder-decoder models is updated based on the result of the joint loss function, wherein the training includes using backpropagation to minimize the reconstruction loss; The ε parameter is used to determine the sphere radius to determine the simplex complex representation of intradomain relations in order to perform the mapping of interdomain relations; as well as Using the updated two or more encoder-decoder models and the sphere radius, a cross-domain structured mapping is determined that correlates dissimilar data of the two or more encoder-decoder models.

18. The apparatus according to claim 17, wherein: The processor is further configured to execute the instructions to: Using the corresponding predictions and the input data from each of the two or more domain corpora of dissimilar data, calculate the corresponding reconstruction loss for each of the two or more encoder-decoder models; and Extract a low-dimensional embedding vector representing the input data from each of the two or more encoder-decoder models; and The two or more domain corpora of dissimilar data include text, images, audio, and other data sources from different knowledge domains.

19. The apparatus according to claim 18, wherein: The processor is further configured to execute the instructions to: Calculate the gradient of the loss from the joint loss function with respect to the model parameters of each of the two or more encoder-decoder models; and The distribution distance metric is the pairwise average relative survival time (MRLT) distribution distance metric, and the function is the joint loss function.

20. The apparatus of claim 19, wherein the processor is further configured to execute the instructions to: Initialize the weights of each of the two or more encoder-decoder models; The input data is preprocessed, transformed, and extracted into feature vectors of fixed dimensions; Perform feedforward processing to feedforward samples of the input data from each domain to each of the two or more encoder-decoder models; Use each of the two or more encoder-decoder models to generate a corresponding output prediction for each of the in-domain samples of the input data; Given the in-domain samples of the input data and the corresponding output prediction, for each of the two or more encoder-decoder models, calculate the corresponding loss value with respect to the joint loss function; The pairwise average relative survival time (MRLT) distribution distance metric is calculated based on a first relative survival time (RLT) matrix and a second relative survival time (RLT) matrix among the samples within each domain of the input data, and based on one or more of the following: The distance between the distributions of these two relative survival time metrics defined between the first and second relative survival time RLT matrices; Use the squared loss function between the output of the first relative survival time RLT matrix and the output of the second relative survival time RLT matrix; or The Wasserstein distance is determined using the distributions of the first relative survival time RLT matrix and the second relative survival time RLT matrix.

Images