skip to main content
US FlagAn official website of the United States government
dot gov icon
Official websites use .gov
A .gov website belongs to an official government organization in the United States.
https lock icon
Secure .gov websites use HTTPS
A lock ( lock ) or https:// means you've safely connected to the .gov website. Share sensitive information only on official, secure websites.

Explore Research Products in the PAR
It may take a few hours for recently added research products to appear in PAR search results.


Title: Interpretable Debiasing of Vectorized Language Representations with Iterative Orthogonalization
We propose a new mechanism to augment a word vector embedding representation that offers improved bias removal while retaining the key information—resulting in improved interpretability of the representation. Rather than removing the information associated with a concept that may induce bias, our proposed method identifies two concept subspaces and makes them orthogonal. The resulting representation has these two concepts uncorrelated. Moreover, because they are orthogonal, one can simply apply a rotation on the basis of the representation so that the resulting subspace corresponds with coordinates. This explicit encoding of concepts to coordinates works because they have been made fully orthogonal, which previous approaches do not achieve. Furthermore, we show that this can be extended to multiple subspaces. As a result, one can choose a subset of concepts to be represented transparently and explicitly, while the others are retained in the mixed but extremely expressive format of the representation.  more » « less
Award ID(s):
2115677
PAR ID:
10465704
Author(s) / Creator(s):
; ; ; ; ; ; ; ; ;
Date Published:
Journal Name:
International Conference on Learning Representations
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ; ; (, Journal of Neurophysiology)
    null (Ed.)
    Perception of limb position and motion combines sensory information from spindles in muscles that span one joint (monoarticulars) and two joints (biarticulars). This anatomical organization should create interactions in estimating limb position. We developed two models, one with only monoarticulars and one with both monoarticulars and biarticulars, to explore how biarticulars influence estimates of arm position in hand ( x, y) and joint ( shoulder, elbow) coordinates. In hand coordinates, both models predicted larger medial-lateral than proximal-distal errors, although the model with both muscle groups predicted that biarticulars would reduce this bias. In contrast, the two models made significantly different predictions in joint coordinates. The model with only monoarticulars predicted that errors would be uniformly distributed because estimates of angles at each joint would be independent. In contrast, the model that included biarticulars predicted that errors would be coupled between the two joints, resulting in smaller errors for combinations of flexion or extension at both joints and larger errors for combinations of flexion at one joint and extension at the other joint. We also carried out two experiments to examine errors made by human subjects during an arm position matching task in which a robot passively moved one arm to different positions and the subjects moved their other arm to mirror-match each position. Errors in hand coordinates were similar to those predicted by both models. Critically, however, errors in joint coordinates were only similar to those predicted by the model with monoarticulars and biarticulars. These results highlight how biarticulars influence perceptual estimates of limb position by helping to minimize medial-lateral errors. NEW & NOTEWORTHY It is unclear how sensory information from muscle spindles located within muscles spanning multiple joints influences perception of body position and motion. We address this issue by comparing errors in estimating limb position made by human subjects with predicted errors made by two musculoskeletal models, one with only monoarticulars and one with both monoarticulars and biarticulars. We provide evidence that biarticulars produce coupling of errors between joints, which help to reduce errors. 
    more » « less
  2. ; ; ; (, IEEE)
    Deep neural networks are susceptible to generating overconfident yet erroneous predictions when presented with data beyond known concepts. This challenge underscores the importance of detecting out-of-distribution (OOD) samples in the open world. In this work, we propose a novel feature-space OOD detection score based on class-specific and class-agnostic information. Specifically, the approach utilizes Whitened Linear Discriminant Analysis to project features into two subspaces the discriminative and residual subspaces - for which the in-distribution (ID) classes are maximally separated and closely clustered, respectively. The OOD score is then determined by combining the deviation from the input data to the ID pattern in both subspaces. The efficacy of our method, named WDiscOOD, is verified on the large-scale ImageNet-1k benchmark, with six OOD datasets that cover a variety of distribution shifts. WDiscOOD demonstrates superior performance on deep classifiers with diverse backbone architectures, including CNN and vision transformer. Furthermore, we also show that WDiscOOD more effectively detects novel concepts in representation spaces trained with contrastive objectives, including supervised contrastive loss and multi-modality contrastive loss. 
    more » « less
  3. ; (, Gender & Society)
    null (Ed.)
    Implicit bias is one of the most successful cases in recent memory of an academic concept being translated into practice. Its use in the National Science Foundation ADVANCE program—which seeks to promote gender equality in STEM (science, technology, engineering, mathematics) careers through institutional transformation—has raised fundamental questions about organizational change. How do advocates translate theories into practice? What makes some concepts more tractable than others? What happens to theories through this translation process? We explore these questions using the ADVANCE program as a case study. Using an inductive, theory-building approach and combination of computational and qualitative methods, we investigate how the concept of implicit bias was translated into practice through the ADVANCE program and identify five key features that made implicit bias useful as a change framework in the academic STEM setting. We find that the concept of implicit bias works programmatically because it is (1) demonstrable, (2) relatable, (3) versatile, (4) actionable, and (5) impartial. While enabling the concept’s diffusion, these characteristics also limit its scope. We reflect on implications for gender theories of organizational change and for practitioners. 
    more » « less
  4. ; ; ; ; ; ; ; ; ; ; et al (, Findings of the Association for Computational Linguistics: ACL 2023)
    Knowledge graph embeddings (KGE) have been extensively studied to embed large-scale relational data for many real-world applications. Existing methods have long ignored the fact many KGs contain two fundamentally different views: high-level ontology-view concepts and fine-grained instance-view entities. They usually embed all nodes as vectors in one latent space. However, a single geometric representation fails to capture the structural differences between two views and lacks probabilistic semantics towards concepts’ granularity. We propose Concept2Box, a novel approach that jointly embeds the two views of a KG using dual geometric representations. We model concepts with box embeddings, which learn the hierarchy structure and complex relations such as overlap and disjoint among them. Box volumes can be interpreted as concepts’ granularity. Different from concepts, we model entities as vectors. To bridge the gap between concept box embeddings and entity vector embeddings, we propose a novel vector-to-box distance metric and learn both embeddings jointly. Experiments on both the public DBpedia KG and a newly-created industrial KG showed the effectiveness of Concept2Box. 
    more » « less
  5. ; (, international conference on machine learning)
    We develop effective approximation methods for unitary matrices. In our formulation, a unitary matrix is represented as a product of rotations in two-dimensional subspaces, so-called Givens rotations. Instead of the quadratic dimension dependence when applying a dense matrix, applying such an approximation scales with the number factors, each of which can be implemented efficiently. Consequently, in settings where an approximation is once computed and then applied many times, such an effective representation becomes advantageous. Although efficient Givens factorizations are not possible for generic unitary operators, we show that minimizing a sparsity-inducing objective with a coordinate descent algorithm on the unitary group yields good factorizations for structured matrices. Canonical applications of such a setup are orthogonal basis transforms. We demonstrate that our methods improve the approximate representation of the graph Fourier transform, the matrix obtained when diagonalizing a graph Laplacian. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email