Search

One challenge in speech translation is that plenty of spoken content is long-form, but short units are necessary for obtaining high-quality translations. To address this mismatch, we adapt large language models (LLMs) to split long ASR transcripts into segments that can be independently translated so as to maximize the overall translation quality. We overcome the tendency of hallucination in LLMs by incorporating finite-state constraints during decoding; these eliminate invalid outputs without requiring additional training. We discover that LLMs are adaptable to transcripts containing ASR errors through prompt-tuning or fine-tuning. Relative to a state-of-the-art automatic punctuation baseline, our best LLM improves the average BLEU by 2.9 points for English-German, English-Spanish, and English-Arabic TED talk translation in 9 test sets, just by improving segmentation., Comment: accepted to the Findings of EMNLP 2023. arXiv admin note: text overlap with arXiv:2212.09895

Ongoing effort has been devoted to applying metamaterials to boost the imaging performance of magnetic resonance imaging owing to their unique capacity for electromagnetic field confinement and enhancement. However, there are still major obstacles to widespread clinical adoption of conventional metamaterials due to several notable restrictions, namely: their typically bulky and rigid structures, deviations in their optimal resonance frequency, and their inevitable interference with the transmission RF field in MRI. Herein, we address these restrictions and report a conformal, smart metamaterial, which may not only be readily tuned to achieve the desired, precise frequency match with MRI by a controlling circuit, but is also capable of selectively amplifying the magnetic field during the RF reception phase by sensing the excitation signal strength passively, thereby remaining off during the RF transmission phase and thereby ensuring its optimal performance when applied to MRI as an additive technology. By addressing a host of current technological challenges, the metamaterial presented herein paves the way toward the wide-ranging utilization of metamaterials in clinical MRI, thereby translating this promising technology to the MRI bedside., Comment: 21 pages, 7 figures

Distributed learning is commonly used for training deep learning models, especially large models. In distributed learning, manual parallelism (MP) methods demand considerable human effort and have limited flexibility. Hence, automatic parallelism (AP) methods have recently been proposed for automating the parallel strategy optimization process. Existing AP methods suffer from sub-optimal solutions because they do not jointly optimize the two categories of parallel strategies (i.e., inter-layer parallelism and intra-layer parallelism). In this paper, we propose a novel AP method called UniAP, which unifies inter- and intra-layer automatic parallelism by mixed integer quadratic programming. To the best of our knowledge, UniAP is the first parallel method that can jointly optimize the two categories of parallel strategies to find an optimal solution. Experimental results show that UniAP outperforms state-of-the-art methods by up to 3.80$\times$ in throughput and reduces strategy optimization time by up to 107$\times$ across five Transformer-based models., Comment: 20 pages, 8 figures

We have constructed 3D Bosons. In this paper, we show the 3D Bosonic Fock space, which is isomorphic to the vector space of 3D Young diagrams as graded vector spaces. We use 3D Bosons to represent the generators of the affine Yangian of ${\mathfrak{gl}}(1)$ and define the 3D cut-and-join operators. Then we discuss the 3D Boson representation of $W$-operators in matrix models., Comment: arXiv admin note: text overlap with arXiv:2212.05665, arXiv:2305.04472, arXiv:2301.04304

Time-periodic form or expression is a ubiquitous natural and man-made phenomenon observable in all the scientific and engineering disciplines. In this article, we propose a theory of periodic sequence (TPS), which can be formulated as a foundational theory for computational sciences and engineering, to transform arbitrary time-periodic electromagnetic (EM) problems into a computational space with mapped discrete events, which is characterized in neither frequency domain nor time domain. Within the TPS framework, periodic-sequential Maxwell's curl equations are decomposed and decoupled to independent and paralleled instances via designated mappings. The fundamental solutions and mapped responses of EM periodic sequences are elucidated, and corroborated by RF/microwave measurements. The nature of outstanding computational parallelism and the unique frequency-independent property make TPS a promising methodology for computational electromagnetics such as analysis of high-speed signal integrity and broadband RF transmission.

We introduce LAST, a LAttice-based Speech Transducer library in JAX. With an emphasis on flexibility, ease-of-use, and scalability, LAST implements differentiable weighted finite state automaton (WFSA) algorithms needed for training \& inference that scale to a large WFSA such as a recognition lattice over the entire utterance. Despite these WFSA algorithms being well-known in the literature, new challenges arise from performance characteristics of modern architectures, and from nuances in automatic differentiation. We describe a suite of generally applicable techniques employed in LAST to address these challenges, and demonstrate their effectiveness with benchmarks on TPUv3 and V100 GPU.

Transaction fee mechanism design is a new decentralized mechanism design problem where users bid for space on the blockchain. Several recent works showed that the transaction fee mechanism design fundamentally departs from classical mechanism design. They then systematically explored the mathematical landscape of this new decentralized mechanism design problem in two settings: in the plain setting where no cryptography is employed, and in a cryptography-assisted setting where the rules of the mechanism are enforced by a multi-party computation protocol. Unfortunately, in both settings, prior works showed that if we want the mechanism to incentivize honest behavior for both users as well as miners (possibly colluding with users), then the miner revenue has to be zero. Although adopting a relaxed, approximate notion of incentive compatibility gets around this zero miner-revenue limitation, the scaling of the miner revenue is nonetheless poor. In this paper, we show that if we make a mildly stronger reasonable-world assumption than prior works, we can circumvent the known limitations on miner revenue, and design auctions that generate optimal miner revenue. We also systematically explore the mathematical landscape of transaction fee mechanism design under the new reasonable-world and demonstrate how such assumptions can alter the feasibility and infeasibility landscape.

Machine learning enables the development of new, supplemental, and empowering tools that can either expand existing technologies or invent new ones. In education, space exists for a tool that supports generic student course review formats to organize and recapitulate students' views on the pedagogical practices to which they are exposed. Often, student opinions are gathered with a general comment section that solicits their feelings towards their courses without polling specifics about course contents. Herein, we show a novel approach to summarizing and organizing students' opinions via analyzing their sentiment towards a course as a function of the language/vocabulary used to convey their opinions about a class and its contents. This analysis is derived from their responses to a general comment section encountered at the end of post-course review surveys. This analysis, accomplished with Python, LaTeX, and Google's Natural Language API, allows for the conversion of unstructured text data into both general and topic-specific sub-reports that convey students' views in a unique, novel way., Comment: 31 Pages, 3 Main Text Figures, 2 Supplemental Figures, 1 Supplements Information Section, 1 Supplemental Sample Report

We construct a two-tensor model with order-3 and present its $W$-representation. Moreover we derive the compact expressions of correlators from the $W$-representation and analyze the free energy in large $N$ limit. In addition, we establish the correspondence between two colored Dyck walks in the Fredkin spin chain and tree operators in the ring. Based on the classification Dyck walks, we give the number of tree operators with the given level. Furthermore, we show the entanglement scaling of Fredkin spin chain beyond logarithmic scaling in the ordinary critical systems from the viewpoint of tensor model., Comment: 28 pages, 15 figures, 2 tables. Revised version accepted for publication in Eur. Phys. J. C

In this paper, we consider 3D Young diagrams with at most $N$ layers in $z$-axis direction, which can be constructed by $N$ 2D Young diagrams on slice $z=j$, $j=1,2,\cdots, N$ from the Yang-Baxter equation. Use 2D Bosons $\{a_{j,m},\ m\in\Z\}$ associated to 2D Young diagrams on the slice $z=j$, we constructed 3D Bosons. Then we show the 3D Boson representation of $W_{1+\infty}$ algebra, and the Littlewood-Richardson rule for 3-Jack polynomials from the actions of 3D Bosons on 3D Young diagrams., Comment: arXiv admin note: text overlap with arXiv:2212.05665

Existing training criteria in automatic speech recognition(ASR) permit the model to freely explore more than one time alignments between the feature and label sequences. In this paper, we use entropy to measure a model's uncertainty, i.e. how it chooses to distribute the probability mass over the set of allowed alignments. Furthermore, we evaluate the effect of entropy regularization in encouraging the model to distribute the probability mass only on a smaller subset of allowed alignments. Experiments show that entropy regularization enables a much simpler decoding method without sacrificing word error rate, and provides better time alignment quality.

3D (3 dimensional) Young diagrams are a generalization of 2D Young diagrams. In this paper, We consider 3D Bosons and 3-Jack polynomials. We associate three parameters $h_1,h_2,h_3$ to $y,x,z$-axis respectively. 3-Jack polynomials are polynomials of $P_{n,j}, n\geq j$ with coefficients in $\mathbb C(h_1,h_2,h_3)$, which are the generalization of Schur functions and Jack polynomials to 3D case. Similar to Schur functions, 3-Jack polynomials can also be determined by the vertex operators and the Pieri formulas.

We present higher Chern-Simons theories based on (2-)crossed modules. We start from the generalized differential forms in Generalized Differential Calculus and define the corresponding generalized connections which consist of higher connections. Then we establish the generalized Chern-Simons forms to get the higher Chern-Simons actions. Finally, we develop the higher second Chern forms and Chern-Weil theorems.

In this paper, we discuss the relations between the Jack polynomials, $\hbar$-dependent KP hierarchy and affine Yangian of ${\mathfrak{gl}}(1)$. We find that $\alpha=\hbar^2$ and $h_1=\hbar, \ h_2=-\hbar^{-1}$, where $\alpha$ is the parameter in Jack polynomials, and $h_1,\ h_2$ are the parameters in affine Yangian of ${\mathfrak{gl}}(1)$. Then the vertex operators which are in Jack polynomials are the same with that in $\hbar$-KP hierarchy, and the Jack polynomials can be used to describe the tau functions of the $\hbar$-KP hierarchy.

Since the ($\beta$-deformed) Hurwitz Kontsevich model corresponds to the special case of affine Yangian of ${\mathfrak{gl}}(1)$. In this paper, we construct two general cases of the $\beta$-deformed Hurwitz Kontsevich model. We find that the $W$-operators of these two models can be represented by the generators $e_k,\ f_k,\psi_k$ of the affine Yangian of ${\mathfrak{gl}}(1)$, and the eigenstates (the symmetric functions $Y_\lambda$ and 3-Jack polynomials) can be obtained from the 3D Young diagram representation of affine Yangian of ${\mathfrak{gl}}(1)$. Then we can see that the $W$-operators and eigenstates are symmetric about the permutations of coordinate axes.

Recent works of Roughgarden (EC'21) and Chung and Shi (SODA'23) initiate the study of a new decentralized mechanism design problem called transaction fee mechanism design (TFM). Unlike the classical mechanism design literature, in the decentralized environment, even the auctioneer (i.e., the miner) can be a strategic player, and it can even collude with a subset of the users facilitated by binding side contracts. Chung and Shi showed two main impossibility results that rule out the existence of a {\it dream} TFM. First, any TFM that provides incentive compatibility for individual users and miner-user coalitions must always have zero miner revenue, no matter whether the block size is finite or infinite. Second, assuming finite block size, no non-trivial TFM can simultaenously provide incentive compatibility for any individual user, and for any miner-user coalition. In this work, we explore what new models and meaningful relaxations can allow us to circumvent the impossibility results of Chung and Shi. Besides today's model that does not employ cryptography, we introduce a new MPC-assisted model where the TFM is implemented by a joint multi-party computation (MPC) protocol among the miners. We prove several feasibility and infeasibility results for achieving {\it strict} and {\it approximate} incentive compatibility, respectively, in the plain model as well as the MPC-assisted model. We show that while cryptography is not a panacea, it indeed allows us to overcome some impossibility results pertaining to the plain model, leading to non-trivial mechanisms with useful guarantees that are otherwise impossible in the plain model. Our work is also the first to characterize the mathematical landscape of transaction fee mechanism design under approximate incentive compatibility, as well as in a cryptography-assisted model.