Showing 1–7 of 7 results for author: Jheng, S

Quantifying Character Similarity with Vision Transformers

Authors: Xinmei Yang, Abhishek Arora, Shao-Yu Jheng, Melissa Dell

Abstract: Record linkage is a bedrock of quantitative social science, as analyses often require linking data from multiple, noisy sources. Off-the-shelf string matching methods are widely used, as they are straightforward and cheap to implement and scale. Not all character substitutions are equally probable, and for some settings there are widely used handcrafted lists denoting which string substitutions ar… ▽ More Record linkage is a bedrock of quantitative social science, as analyses often require linking data from multiple, noisy sources. Off-the-shelf string matching methods are widely used, as they are straightforward and cheap to implement and scale. Not all character substitutions are equally probable, and for some settings there are widely used handcrafted lists denoting which string substitutions are more likely, that improve the accuracy of string matching. However, such lists do not exist for many settings, skewing research with linked datasets towards a few high-resource contexts that are not representative of the diversity of human societies. This study develops an extensible way to measure character substitution costs for OCR'ed documents, by employing large-scale self-supervised training of vision transformers (ViT) with augmented digital fonts. For each language written with the CJK script, we contrastively learn a metric space where different augmentations of the same character are represented nearby. In this space, homoglyphic characters - those with similar appearance such as ``O'' and ``0'' - have similar vector representations. Using the cosine distance between characters' representations as the substitution cost in an edit distance matching algorithm significantly improves record linkage compared to other widely used string matching methods, as OCR errors tend to be homoglyphic in nature. Homoglyphs can plausibly capture character visual similarity across any script, including low-resource settings. We illustrate this by creating homoglyph sets for 3,000 year old ancient Chinese characters, which are highly pictorial. Fascinatingly, a ViT is able to capture relationships in how different abstract concepts were conceptualized by ancient societies, that have been noted in the archaeological literature. △ Less

Submitted 23 May, 2023; originally announced May 2023.

Linking Representations with Multimodal Contrastive Learning

Authors: Abhishek Arora, Xinmei Yang, Shao-Yu Jheng, Melissa Dell

Abstract: Many applications require linking individuals, firms, or locations across datasets. Most widely used methods, especially in social science, do not employ deep learning, with record linkage commonly approached using string matching techniques. Moreover, existing methods do not exploit the inherently multimodal nature of documents. In historical record linkage applications, documents are typically n… ▽ More Many applications require linking individuals, firms, or locations across datasets. Most widely used methods, especially in social science, do not employ deep learning, with record linkage commonly approached using string matching techniques. Moreover, existing methods do not exploit the inherently multimodal nature of documents. In historical record linkage applications, documents are typically noisily transcribed by optical character recognition (OCR). Linkage with just OCR'ed texts may fail due to noise, whereas linkage with just image crops may also fail because vision models lack language understanding (e.g., of abbreviations or other different ways of writing firm names). To leverage multimodal learning, this study develops CLIPPINGS (Contrastively LInking Pooled Pre-trained Embeddings). CLIPPINGS aligns symmetric vision and language bi-encoders, through contrastive language-image pre-training on document images and their corresponding OCR'ed texts. It then contrastively learns a metric space where the pooled image-text embedding for a given instance is close to embeddings in the same class (e.g., the same firm or location) and distant from embeddings of a different class. Data are linked by treating linkage as a nearest neighbor retrieval problem with the multimodal embeddings. CLIPPINGS outperforms widely used string matching methods by a wide margin in linking mid-20th century Japanese firms across financial documents. A purely self-supervised model - trained only by aligning the embeddings for the image crop of a firm name and its corresponding OCR'ed text - also outperforms popular string matching methods. Fascinatingly, a multimodally pre-trained vision-only encoder outperforms a unimodally pre-trained vision-only encoder, illustrating the power of multimodal pre-training even if only one modality is available for linking at inference time. △ Less

Submitted 21 June, 2024; v1 submitted 6 April, 2023; originally announced April 2023.

A simple method to construct eigenset of single-active-electron atom in momentum space with applications to solve time-dependent Schroedinger equation

Authors: Shih-Da Jheng, Tsin-Fu Jiang

Abstract: We present a highly accurate method for solving single-active-electron (SAE) atomic eigenset in momentum space. The trouble of Coulomb kernel singularity is bypassed with numerical quadrature, which is simple but effective. The complicated Lande regularization method is no longer necessary. The data of accuracy for some low-lying states of the hydrogen and SAE helium atom were tabulated. Two examp… ▽ More We present a highly accurate method for solving single-active-electron (SAE) atomic eigenset in momentum space. The trouble of Coulomb kernel singularity is bypassed with numerical quadrature, which is simple but effective. The complicated Lande regularization method is no longer necessary. The data of accuracy for some low-lying states of the hydrogen and SAE helium atom were tabulated. Two examples of using the generated eigenset to solve the hydrogen atom under strong-field laser pulses were shown. The momentum and the coordinate representation are complementary to each other in quantum mechanics. The simple method to generate eigenstates and the localized behavior of wave functions in momentum space would be useful in the study of quantum mechanical problems involving continuous states. △ Less

Submitted 17 June, 2017; originally announced June 2017.

Comments: 15 pages, 3 tables, 2 figures

Physical Realization of von Neumann Lattices in Rotating Dipole-blockaded Bose Gases

Authors: Szu-Cheng Cheng, Shih-Da Jheng

Abstract: A mathematical lattice, called the von Neumann lattice, is a subset of coherent states and exists periodically in the phase space. It is unlike solids or Abrikosov lattices that are observable in physical systems. Abrikosov lattices are vortices closely packed into a lattice with a flux quantum through a unit cell. Although Abrikosov lattices appear generally in various physical systems, vortex la… ▽ More A mathematical lattice, called the von Neumann lattice, is a subset of coherent states and exists periodically in the phase space. It is unlike solids or Abrikosov lattices that are observable in physical systems. Abrikosov lattices are vortices closely packed into a lattice with a flux quantum through a unit cell. Although Abrikosov lattices appear generally in various physical systems, vortex lattices with multiple-flux quantums through a unit cell are more stable than Abrikosov lattices in some physical regimes of the systems with non-local interactions between particles. No theory is able to describe these vortex lattices today. Here, we develop a theory for these vortex lattices by extending von Neumann lattices to the coordinate space with a unit cell of area that is proportional to flux quantums through a unit cell. The von Neumann lattices not only show the same physical properties as the Abrikosov lattice, but also describe vortex lattices with multiple-flux quantums through a unit cell. From numerical simulations of a rapidly rotating dipole-blockaded gas, we confirm that vortex lattices showed in our simulations are the representation of von Neumann lattices in the coordinate space. We anticipate our theory to be a starting point for develo** more sophisticated vortex-lattice models. For example, the effect of Landau-level mixing on vortex lattice structures, vortices formed inside superfluid droplets and structural phase transitions of vortex matter in two-component Bose-Einstein condensates will be relevant for such developments. △ Less

Submitted 8 April, 2015; originally announced April 2015.

Comments: 22 pages, 5 figures and 1 table

Roton Instabilities and Wigner Crystallization of Rotating Dipolar Fermions in the Fractional Quantum Hall Regime

Authors: Shih-Da Jheng, T. F. Jiang, Szu-Cheng Cheng

Abstract: We point out the possibility of occurring instabilities in Laughlin liquids of rotating dipolar fermions with zero thickness. Previously such a system was predicted to be the Laughlin liquid for filling factors being greater and equal to 1/7. However, from intra-Landau-level excitations of the liquid in the single-mode approximation, the roton minima become negative and Laughlin liquids are unstab… ▽ More We point out the possibility of occurring instabilities in Laughlin liquids of rotating dipolar fermions with zero thickness. Previously such a system was predicted to be the Laughlin liquid for filling factors being greater and equal to 1/7. However, from intra-Landau-level excitations of the liquid in the single-mode approximation, the roton minima become negative and Laughlin liquids are unstable for filling factors being less and equal to 1/7. We then conclude that there are correlated Wigner crystals for filling factors being less and equal to 1/7. △ Less

Submitted 15 May, 2013; originally announced May 2013.

Comments: 4 pages, 4 figures. arXiv admin note: substantial text overlap with arXiv:0808.0439

Journal ref: Phys. Rev. A 88, 051601 (2013)

Wigner Crystallization of Rotating Dipolar Fermions in the Fractional Quantum Hall Regime

Authors: Szu-Cheng Cheng, Shih-Da Jheng, T. F. Jiang

Abstract: We show the possible existence of the Wigner crystal (WC) in the Fractional Quantum Hall (FQH) regime. We find that the Landau-level mixing (LLM) will lower the energy of the WC significantly in the high-density regime. The WC is lower in energy than the FQH liquid in the high-density regime. We conclude that the crystal phase is expected at high density for rotating dipolar gases, which is cons… ▽ More We show the possible existence of the Wigner crystal (WC) in the Fractional Quantum Hall (FQH) regime. We find that the Landau-level mixing (LLM) will lower the energy of the WC significantly in the high-density regime. The WC is lower in energy than the FQH liquid in the high-density regime. We conclude that the crystal phase is expected at high density for rotating dipolar gases, which is consistent with non-rotating dipolar gases, but is inconsistent with the low-density conclusion from Baranov et al. [Phys. Rev. Lett. 100, 200402 (2008)], where the effect of LLM is ignored. △ Less

Submitted 10 May, 2010; v1 submitted 18 April, 2010; originally announced April 2010.

Comments: 15 pages, 3 figures

Quantum Melting of a Wigner crystal of Rotating Dipolar Fermions in the Lowest Landau Level

Authors: Szu-Cheng Cheng, Shih-Da Jheng, T. F. Jiang

Abstract: We have investigated the behavior and stability of a Wigner crystal of rotating dipolar fermions in two dimensions. Using an ansatz wave function for the ground state of rotating two-dimensional dipolar fermions, which occupy only partially the lowest Landau level, we study the correlation energy, elastic moduli and collective modes of Wigner crystals in the lowest Landau level. We then calculat… ▽ More We have investigated the behavior and stability of a Wigner crystal of rotating dipolar fermions in two dimensions. Using an ansatz wave function for the ground state of rotating two-dimensional dipolar fermions, which occupy only partially the lowest Landau level, we study the correlation energy, elastic moduli and collective modes of Wigner crystals in the lowest Landau level. We then calculate the mean square of the displacement vector of Wigner crystals. The critical filling factor, below which the crystalline state is expected, is evaluated at absolute zero by use of the Lindeman's criterion. We find that the particle (hole) crystal is locally stable for filling factor is less than 1/15 (between filling factors 14/15 and 1), where the stable regime of the crystal is much narrower than the result from Baranov, Fehrmann and Lewenstein, [Phys. Rev. Lett. 100, 200402 (2008)]. △ Less

Submitted 14 March, 2010; v1 submitted 10 March, 2010; originally announced March 2010.

Comments: 28 pages and 9 figures