: o_i∈

}. Measuring task performance for imputation by the completion pipeline 300 can involve computing the Root Mean Square Error (RMSE). For numerical features, the RMSE can be calculated as:

$\begin{matrix} RMSE (X_{test}) = \frac{1}{N_{test}} \sum_{i = 1}^{N_{test}} \sqrt{\frac{\sum_{j = 1}^{M_{i}} {({\hat{x}}_{i}^{j} - y_{i}^{j})}^{2}}{M_{i}}} . & Equation (2) \end{matrix}$

Here, N_testis the number of test samples and M_iis missing numerical features for sample i.

${\hat{x}}_{i}^{j}$

is imputation prediction for missing feature j of sample i by the completion pipeline 300. Furthermore,

$y_{i}^{j}$

is the dataset value for missing feature j of sample i and represents a target design for a testing sample.

Comparing model performance in imputing categorical features can involve computing error rates as:

$\begin{matrix} Err (X_{test}) = \frac{1}{N_{test}} \frac{1}{M_{i}} \sum_{i = 1}^{N_{test}} \sum_{j = 1}^{M_{i}} [{\hat{x}}_{i}^{j} \neq y_{i}^{j}] . & Equation (3) \end{matrix}$

Here, N_testis the number of test samples. M_iis the number of missing categorical features for sample i and [Image Omitted] is an indicator function. Furthermore,

$y_{i}^{j}$

is the dataset value for missing feature j of sample i and represents a target design for a testing sample.

Diversity of outputs imputed and generated by the completion pipeline 300 can be measured through (a) a diversity score to measure coverage, and (b) Kullback-Leibler (KL)-Divergence from distributions of generated features when compared to a dataset that measures data representativeness. For example, the diversity score is computed as:

$\begin{matrix} Diversity (X_{test}) = \frac{1}{N_{test}} \frac{1}{M_{i}} \sum_{i = 1}^{N_{test}} \sum_{j = 1}^{M_{i}} \max_{k \neq l} \sqrt{{(S_{k, j} - S_{l, j})}^{2}} . & Equation (4) \end{matrix}$

Here, l and k are the index of the samples obtained by the generative model and S_k,jand S_l,jare the k-th and jl-th sampled value for missing feature j. The diversity score finds the average maximum distance between missing features, parameter gaps, missing parameters, etc. The diversity score factors that some missing features can exhibit constrained correct values while others have a large variation. In one approach, the estimation system 100 considers features that have a mean correlation value that is greater than the median among features of the dataset that are primarily factored when calculating the diversity score.

Still referring to FIG. 3, the completion pipeline 300 can accept the partial parameters 150 and an assembly graph about an object from a computing task (e.g., design, ideation, testing, etc.). The completion pipeline 300 constructs a parameter graph 310 having details that accurately describe features and interdependencies between features for the object. The estimation system 100 can derive these insights from the assembly graph conditioned with the partial parameters 150. For example, the object is a bike and the computing task is designing a new bike. Here, the partial parameters 150 can describe features of the bike (e.g., color, mountain, etc.) and the assembly graph describes couplings between different components of the bike (e.g., handlebar, wheels, seat, etc.). Furthermore, the generation module 130 can create a graph embedding from encoding the parameter graph 310 using a learning model 340 (e.g., a GNN, a GCN, etc.). Here, the learning model 340 can train by minimizing losses such that the graph embedding includes nuanced features about the object and accurately represents the structural relationships between object components. In this way, the completion pipeline 300 efficiently executes the computing task through a structure and organization of the graph embedding.

Moreover, a feature tokenizer 320 and a positional encoder 330 translate parametric and spatial information into a tokenized format representing embeddings. Here, the feature tokenizer 320, the positional encoder 330, and the learning model 340 can execute information in parallel, serially, etc., such as according to the complexity of the computing task. In one approach, the feature tokenizer 320 derives a parametric embedding from the assembly graph. An embedding can be one of a vector (e.g., flow tensors) and a number array that represent features about the object using various values (e.g., −1 to 1) and a simple format. Furthermore, the positional encoder 330 computes a positional embedding from the assembly graph. The positional embedding can include positional information of features within the object and relational context about object features. For instance, features about a handlebar positioned proximate to a bike frame include related information (e.g., size, connection type, etc.).

As further explained below, a cross-attention model 350 can estimate a conditional embedding that is multi-modal using the graph embedding, the parametric embedding, and the positional embedding. A cross-attention network can be a neural network processing multiple sequences of data for identifying relevant relationships and focus areas. In one approach, an input is a “query” sequence (e.g., a textual input) and another sequence is a “key” sequence defining context, additional information, etc. By selectively focusing on relevant parts of the key sequence while processing elements of the query sequence, the network generates more accurate and contextually important outputs. Here, the cross-attention model 350 selects structural context and information about the object that is relevant within the embeddings. As such, the cross-attention model 350 can intelligently focus upon and fuse applicable data for other computing tasks to complete the partial parameters 150. The output of the cross-attention model 350 is a conditional embedding that is multi-modal through incorporating different input categories and types. The output can be conditional since the completion pipeline 300 begins processes using existing information with the partial parameters 150 and the assembly graph rather than random data.

Additionally, the completion pipeline 300 can also include a denoising model 360 (e.g., diffusion-based denoising) to accurately impute the partial parameters 150 from the multi-modal conditional embedding. As previously explained, denoising can using forward-diffusion for injecting noise into the partial parameters 150 and the assembly graph completing the partial parameters 150. Here, the denoising model 360 can extract features from the partial parameters 150 and the assembly graph having injected noise. As previously explained, a network can interpret outputs from the denoising model 360 for completing the partial parameters 150. Furthermore, the completion pipeline 300 can include the rending engine 370 (e.g., a computer-aided design (CAD) engine) that generates various objects having diverse properties using completed parameters. The parameters for the objects generated by the completion pipeline 300 can be further modified upon evaluation 380 leading to additional gaps, missing values, etc., for the partial parameters 150. In other words, the evaluation 380 is feedback for iterations and refinement of generated objects. Accordingly, the completion pipeline 300 facilitates the generation of accurate and detailed objects that bridge the gap between partial information and complete realization.

Regarding details about constructing the parameter graph 310 from an assembly graph and the partial parameters 150, the completion pipeline 300 may categorize features. For example, the completion pipeline 300 categorizes features into 11 distinct components in Table 1 for a bike. The completion pipeline 300 can translate the components in Table 1 into a tabular form that simplifies computations. Although Table 1 references a bike, the completion pipeline 300 can complete the partial parameters 150 for any object during a computing task as understood by one of ordinary skill in the art.

TABLE 1

Component
Features

Seat Tube
Seat tube length, Seat tube extension2, Seat tube diameter, Seatpost setback,

Seatpost LENGTH, Stack, Lower stack height, Upper stack height

Head Tube
Head tube upper extension2, Head angle, Head tube lower extension2, Head

tube diameter

Top Tube
Top tube rear diameter, Top tube front diameter

Down
Down tube front diameter, Down tube rear diameter

Tube

Chain Stay
CHAINSTAYAUXrearDIAMETER, Chain stay horizontal diameter, Chain

stay position on BB, Chain stay taper, Chain stay back diameter, Chainstay

vertical diameter

Seat Stay
Seat stay junction0, Seat stay bottom diameter, SEATSTAY_HF,

SEATSTAY_HR, SEATSTAYTAPERLENGTH

Fork
FORK10R, FORK0L

Saddle
Saddle P, Saddle thickness, Saddle angle, Saddle J, Saddle H, Saddle E,

SADDLETIPtoMIDDLE, Saddle length

Wheel
Wheel width rear, Wheel width front, Wheel diameter front, Wheel diameter

rear, SPOKES composite front, SPOKES front, SPOKES rear, SPOKES

composite rear, ERD rear, ERD front

Handle
Road bar reach, Road bar drop, Brake lever position, Bullhorn angle,

HBAREXTEND, Handlebar angle, MtnBar angle, HBARTHETA, Pedal

width, Stem angle, Stem length

BB
BB textfield, BB length, BB diameter

Now referring to FIGS. 4A and 4B, examples of classifications for an assembly graph 400 and automatically completing partial designs are illustrated. The assembly graph 400 delineates a structural and relational framework of the bike components. For example, a saddle is structurally coupled to a seat tube but not the seat stay. However, the seat tube is structurally coupled to the seat stay and the top tube. The completion pipeline 300 can utilize the partial parameters 150 and the assembly graph 400 for systematically constructing the parameter graph 310 in a feature-specific manner. This allows the learning model 340 (e.g., a GNN, GCN, etc.) to output encodings of object features that are context-aware through interconnections and architecture represented within the parameter graph 310. In this way, the completion pipeline 300 ensures that the learning model 340 has a detailed understanding of object structure that facilitates accurate feature imputation and encoding.

Regarding graphing structure, node features within the parameter graph 310 can be concatenated features for components within the assembly graph 400 and modified as specified by the partial parameters 150. In one approach, missing features, parameters, etc., have a value of 0 that preserve feature sizes. An edge that is weighted exists between nodes when corresponding components are physically coupled, interact, etc., within a bike structure. For instance, a weight having 0 value indicates unrelated components while a weight having 1 value indicates related components. As such, the assembly graph 400 delineates component relationships and allows the learning model 340 to process partial parameters 150 while maintaining natural connectivity and dependency between different parts. Thus, the learning model 340 accurately captures and encodes complex interplay of components, thereby facilitating completed parameters across distributions that are more nuanced and informed during generative computations.

Regarding details about estimating a conditional embedding of a graph embedding and the assembly graph 400, the feature tokenizer 320 may have fully connected layers that process and encode numerical features and output the parametric embedding. In one approach, the feature tokenizer 320 has an embedding layer that is dedicated and implemented for the categorical features so that each a feature type is optimally represented. Furthermore, the positional encoder 330 captures spatial aspects of the assembly graph 400 through encoding positional information about features. In this way, the positional encoder 330 focuses upon relational context found in tabular data such that features positioned closely are likely to convey related information. Furthermore, the cross-attention model 350 may fuse embeddings into a multi-model representation using an embedding derived from the assembly graph 400 with those generated by the feature tokenizer 320 and the positional encoder 330. As such, the cross-attention model 350 effectively integrates structural, categorical, and spatial dimensions of data into a conditional embedding that is singular and multi-modal. The integrated embeddings capture both intrinsic properties and interrelations of features within the assembly graph 400, thereby improving diffusion and denoising for completion.

In one embodiment, the denoising model 360 completes the partial parameters 150 using the conditional embedding. Here, the conditional embedding can represent a multi-modal embedding vector enriched with the assembly graph 400, a parametric embedding, and a positional embedding. The denoising model 360 can automatically generate the missing parameters through a diffusion operation that ensures both diversity and accuracy of completed objects.

Moreover, FIG. 4B shows visualization from a rendering pipeline that transforms the object into various forms using a parametric set that is completed and outputted from the completion pipeline 300. Here, the partial designs 4101 of speed bikes having the partial parameters 150 are automatically completed as designs 4201 exhibiting various wheel types. The designs 4201 exhibit diverse variations for wheel types generated from graph-based learning, multi-modal attention networks, and the denoising model 360. Similarly, the partial designs 4102 of mountain bikes having the partial parameters 150 are automatically completed as designs 4202 that have diverse variations and life-like properties. The designs 4202 have multi-faceted outputs for seats and handlebars. Accordingly, the completion pipeline 300 mimics human design through outputting completed objects along precise partial parameters while having individualized qualities.

Now discussing FIG. 5, one embodiment of a method that is associated with estimating a conditional embedding from partial parameters and an assembly graph using a cross-attention model and automatically completing parameters using a denoising model is illustrated. Method 500 will be discussed from the perspective of the estimation system 100 of FIG. 1. While the method 500 is discussed in combination with the estimation system 100, it should be appreciated that the method 500 is not limited to being implemented within the estimation system 100 but is instead one example of a system that may implement the method 500.

The method 500 accurately imputes missing data and generates objects that are diverse and realistic using the estimation system 100. Furthermore, the diffusion of embeddings derived from the partial parameters 150 and an assembly graph promotes the exploration and creativity of atypical objects through spreading information across multi-domains using injected noise. In one approach, the method 500 also generates coherent objects from weakly correlated features and relational cues that are minimal from the partial parameters