An official website of the United States government
Large-scale general domain pretraining followed by downstream-specific finetuning has become a predominant paradigm in machine learning. However, discrepancies between the pretraining and target domains can still lead to performance degradation in certain cases, underscoring the need for task-adaptive continued pretraining (TAP). TAP methods typically involve continued pretraining on task-specific unlabeled datasets or introducing additional unsupervised learning objectives to enhance model capabilities. While many TAP methods perform continued pretraining with multiple pretraining objectives, they often determine the tradeoff parameters between objectives manually, resulting in suboptimal outcomes and higher computational costs. In this paper, we propose TapWeight, a task-adaptive pretraining framework which automatically determines the optimal importance of each pretraining objective based on downstream feedback. TapWeight reweights each pretraining objective by solving a multi-level optimization problem. We applied TapWeight to both molecular property prediction and natural language processing tasks, significantly surpassing baseline methods. Experimental results validate the effectiveness and generalizability of TapWeight.
more »
« less
Implementing dynamic subset sensitivity analysis for early design datasets
Engaging with performance feedback in early building design often involves building a custom parametric model and generating large datasets, which is not always feasible. Alternatively, large parametric datasets of general design problems and filtering methods could be used together to explore specific design decisions. This paper investigates the generalizability of a method that dynamically assesses variable importance and likely influence on performance objectives as a precomputed design space is filtered down. The method first trains linear model trees to predict building performance objectives across a generic design space. Leaf node models are then aggregated to provide feedback on variable importance in different design space regions. This approach is tested on three design problems that vary in number of variables, samples, and design space structure to reveal advantages and potential limitations of the method. Algorithm improvements are proposed, and general recommendations are developed to apply it on future datasets.
more »
« less
- Award ID(s):
- 2033332
- PAR ID:
- 10512297
- Publisher / Repository:
- ScienceDirect
- Date Published:
- Journal Name:
- Automation in Construction
- Volume:
- 158
- Issue:
- C
- ISSN:
- 0926-5805
- Page Range / eLocation ID:
- 105198
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Abstract Recovery‐based design links building‐level engineering and broader community resilience objectives. However, the relationship between above‐code engineering improvements and recovery performance is highly nonlinear and varies on a building‐ and site‐specific basis, presenting a challenge to both individual owners and code developers. In addition, downtime simulations are computationally expensive and hinder exploration of the full design space. In this paper, we present an optimization framework to identify optimal above‐code design improvements to achieve building‐specific recovery objectives. We supplement the optimization with a workflow to develop surrogate models that (i) rapidly estimate recovery performance under a range of user‐defined improvements, and (ii) enable complex and informative optimization techniques that can be repeated for different stakeholder priorities. We explore the implementation of the framework using a case study office building, with a 50th percentile baseline functional recovery time of 155 days at the 475‐year ground‐motion return period. To optimally achieve a target recovery time of 21 days, we find that nonstructural component enhancements are required, and that increasing structural strength (through increase of the importance factor) can be detrimental. However, for less ambitious target recovery times, we find that the use of larger importance factors eliminates the need for nonstructural component improvements. Such results demonstrate that the relative efficacy of a given recovery‐based design strategy will depend strongly on the design criteria set by the user.more » « less
-
Abstract Nonlinear response history analysis (NLRHA) is generally considered to be a reliable and robust method to assess the seismic performance of buildings under strong ground motions. While NLRHA is fairly straightforward to evaluate individual structures for a select set of ground motions at a specific building site, it becomes less practical for performing large numbers of analyses to evaluate either (1) multiple models of alternative design realizations with a site‐specific set of ground motions, or (2) individual archetype building models at multiple sites with multiple sets of ground motions. In this regard, surrogate models offer an alternative to running repeated NLRHAs for variable design realizations or ground motions. In this paper, a recently developed surrogate modeling technique, called probabilistic learning on manifolds (PLoM), is presented to estimate structural seismic response. Essentially, the PLoM method provides an efficient stochastic model to develop mappings between random variables, which can then be used to efficiently estimate the structural responses for systems with variations in design/modeling parameters or ground motion characteristics. The PLoM algorithm is introduced and then used in two case studies of 12‐story buildings for estimating probability distributions of structural responses. The first example focuses on the mapping between variable design parameters of a multidegree‐of‐freedom analysis model and its peak story drift and acceleration responses. The second example applies the PLoM technique to estimate structural responses for variations in site‐specific ground motion characteristics. In both examples, training data sets are generated for orthogonal input parameter grids, and test data sets are developed for input parameters with prescribed statistical distributions. Validation studies are performed to examine the accuracy and efficiency of the PLoM models. Overall, both examples show good agreement between the PLoM model estimates and verification data sets. Moreover, in contrast to other common surrogate modeling techniques, the PLoM model is able to preserve correlation structure between peak responses. Parametric studies are conducted to understand the influence of different PLoM tuning parameters on its prediction accuracy.more » « less
-
Abstract Compositionally graded alloys, a subclass of functionally graded materials (FGMs), utilize localized variations in composition with a single metal part to achieve higher performance than traditional single material parts. In previous work [Kirk, T., Galvan, E., Malak, R., and Arroyave, R., 2018, “Computational Design of Gradient Paths in Additively Manufactured Functionally Graded Materials,” J. Mech. Des., 140, p. 111410. 10.1115/1.4040816], the authors presented a computational design methodology that avoids common issues which limit a gradient alloy’s feasibility, such as deleterious phases, and optimizes for performance objectives. However, the previous methodology only samples the interior of a composition space, meaning designed gradients must include all elements in the space throughout the gradient. Because even small amounts of additional alloying elements can introduce new deleterious phases, this characteristic often neglects potentially simpler solutions to otherwise unsolvable problems and, consequently, discourages the addition of new elements to the state space. The present work improves upon the previous methodology by introducing a sampling method that includes subspaces with fewer elements in the design search. The new method samples within an artificially expanded form of the state space and projects samples outside the true region to the nearest true subspace. This method is evaluated first by observing the sample distribution in each subspace of a 3D, 4D, and 5D state space. Next, a parametric study in a synthetic 3D problem compares the performance of the new sampling scheme to the previous methodology. Lastly, the updated methodology is applied to design a gradient from stainless steel to equiatomic NiTi that has practical uses such as embedded shape memory actuation and for which the previous methodology fails to find a feasible path.more » « less
-
Parametric optimization techniques allow building designers to pursue multiple performance objectives, which can benefit the overall design. However, the strategies used by architecture and engineering graduate students when working with optimization tools are unclear, and ineffective computational design procedures may limit their success as future designers. In response, this re-search identifies several designerly behaviors of graduate students when responding to a conceptual building design optimization task. It uses eye-tracking, screen recording, and empirical methods to code their behaviors following the situated FBS framework. From these data streams, three different types of design iterations emerge: one by the designer alone, one by the optimizer alone, and one by the designer incorporating feedback from the optimizer. Based on the timing and frequency of these loops, student participants were characterized as completing partial, crude, or complete optimization cycles while developing their designs. This organization of optimization techniques establishes reoccurring strategies employed by developing designers, which can encourage future pedagogical approaches that empower students to incorporate complete optimization cycles while improving their designs. It can also be used in future research studies to establish clear links between types of design optimization behavior and design quality.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Journal Article:
- https://doi.org/10.1016/j.autcon.2023.105198
