arXiv subjects

Zichao Yang

Publications and source records attributed to Zichao Yang.

At least 19 recordsLinked to original sources

arXiv2026

Partial Identification of Spatial Production Networks

Which regional exposure conclusions are identified when public data do not observe buyer-seller links across states? We study this question by treating the missing intermediate-input spatial kernel as an unknown coupling constrained by regional activity margins, support restrictions, and auxiliary shipment moments. For linear exposure statistics, the sharp identified set is computed by transportation linear programs. Applying the method to U.S. state-sector data, we find that shipment data are inconsistent with the spatial diffuseness implied by proportional regionalization in key goods sectors. However, they do not identify a unique regional production network or a precise ranking of state exposure to local shocks. Bilateral shipment restrictions tighten the bounds, but much of the remaining uncertainty comes from large service and mixed sectors that are weakly covered by goods-movement data. The results show which exposure conclusions are supported by public data and which are imposed by maintained regionalization assumptions.

econ.GN↗

arXiv2026

The Anatomy of a Blockchain Prediction Market: Polymarket in the 2024 U.S. Presidential Election

Using on-chain Polygon data, we analyze Polymarket's 2024 U.S. Presidential Election market and develop a transaction-level accounting framework with two components: a volume decomposition that separates exchange-equivalent turnover from share minting and burning, and trader-level disagreement measures. Naive aggregation reports $958M of October Trump-market volume, compared with $391M under our decomposition. Market quality improved as arbitrage-deviation half-lives fell from hours to under a minute and Kyle's {\lambda} dropped from 0.53 to 0.01. During October's large-account episode, capital flowed into both sides simultaneously, consistent with heterogeneous-beliefs trading rather than one-sided manipulation. The framework generalizes to other tokenized prediction markets.

econ.GN↗

arXiv2026

Political Shocks and Price Discovery in Prediction Markets: Evidence from the 2024 U.S. Presidential Election

What do trading and prices each reveal when political news hits a prediction market? We answer using Polymarket's on-chain ledger around three shocks in the 2024 U.S. presidential election: the Biden-Trump debate, the assassination attempt on Trump, and Biden's withdrawal. Trading rises after every shock, mainly among incumbents with greater prior activity and with larger realized gains on pre-event portfolios. Price adjustment differs across shocks. The debate's price increase largely reverses, the assassination-attempt repricing persists, and Biden's withdrawal generates the heaviest trading with little change in the Trump price. The price response tracks what the news reveals about linked candidates and how much was already anticipated, not the amount of trading. Trading volume measures participation, while belief revision must be read from linked outcome prices and the surprise in the news.

econ.GN↗

arXiv2025

Waiting Time Distribution of Supra-Arcade Downflows

Supra-arcade downflows (SADs) are dark, sunward-moving structures above the arcade of flare loops. Naturally they are considered to be associated with outflows resulting from magnetic reconnection at the vertical current sheet beneath the eruptive structure. Despite the extensive investigations on the SAD properties, the timing information, particularly the waiting time between consecutive SADs, has not been systematically examined. Here, using the 131~{\AA} passband of the Atmospheric Imaging Assembly on-board the Solar Dynamics Observatory, we studied the waiting time distribution (WTD) of SADs in 7 eruptive flares. In six of the 7 flares, the SADs are identified and tracked in previous studies, by two different methods; and in the 7th flare, by our optimized manual method. Based on statistical evaluations, we found that most of the WTDs are best fitted by a power-law function with the slope ranging from 1.7 to 2.4, in accordance with the prediction of non-stationary Poisson processes or self-organized criticality; but often they can also be fitted almost equally well by a log-normal function. These results rule out linear random or quasi-periodic processes to be responsible for the generation of SADs, but suggest that several nonlinear mechanisms be coupled together in the reconnection outflow region to shape the heavy-tailed WTD of SADs.

astro-ph.SR↗

arXiv2025

ESG Rating Disagreement and Corporate Total Factor Productivity:Inference and Prediction

This paper examines how ESG rating disagreement (Dis) affects corporate total factor productivity (TFP) in China based on data of A-share listed companies from 2015 to 2022. We find that Dis reduces TFP, especially in state-owned, non-capital-intensive, low-pollution and high-tech firms, green innovation strengthens the dampening effect of Dis on TFP, and that Dis lowers corporate TFP by increasing financing constraints and weakening human capital. Furthermore, XGBoost regression demonstrates that Dis plays a significant role in predicting TFP, with SHAP showing that the dampening effect of ESG rating disagreement on TFP is still pronounced in firms with large Dis values.

econ.GN↗

arXiv2024

Unified Generation, Reconstruction, and Representation: Generalized Diffusion with Adaptive Latent Encoding-Decoding

The vast applications of deep generative models are anchored in three core capabilities -- generating new instances, reconstructing inputs, and learning compact representations -- across various data types, such as discrete text/protein sequences and continuous images. Existing model families, like variational autoencoders (VAEs), generative adversarial networks (GANs), autoregressive models, and (latent) diffusion models, generally excel in specific capabilities and data types but fall short in others. We introduce Generalized Encoding-Decoding Diffusion Probabilistic Models (EDDPMs) which integrate the core capabilities for broad applicability and enhanced performance. EDDPMs generalize the Gaussian noising-denoising in standard diffusion by introducing parameterized encoding-decoding. Crucially, EDDPMs are compatible with the well-established diffusion model objective and training recipes, allowing effective learning of the encoder-decoder parameters jointly with diffusion. By choosing appropriate encoder/decoder (e.g., large language models), EDDPMs naturally apply to different data types. Extensive experiments on text, proteins, and images demonstrate the flexibility to handle diverse data and tasks and the strong improvement over various existing models.

cs.LG↗

arXiv2023

Composable Text Controls in Latent Space with ODEs

Real-world text applications often involve composing a wide range of text control operations, such as editing the text w.r.t. an attribute, manipulating keywords and structure, and generating new text of desired properties. Prior work typically learns/finetunes a language model (LM) to perform individual or specific subsets of operations. Recent research has studied combining operations in a plug-and-play manner, often with costly search or optimization in the complex sequence space. This paper proposes a new efficient approach for composable text operations in the compact latent space of text. The low-dimensionality and differentiability of the text latent vector allow us to develop an efficient sampler based on ordinary differential equations (ODEs) given arbitrary plug-in operators (e.g., attribute classifiers). By connecting pretrained LMs (e.g., GPT2) to the latent space through efficient adaption, we then decode the sampled vectors into desired text sequences. The flexible approach permits diverse control operators (sentiment, tense, formality, keywords, etc.) acquired using any relevant data from different domains. Experiments show that composing those operators within our approach manages to generate or edit high-quality text, substantially improving over previous methods in terms of generation quality and efficiency.

cs.CL↗

arXiv2023

Language Models Meet World Models: Embodied Experiences Enhance Language Models

While large language models (LMs) have shown remarkable capabilities across numerous tasks, they often struggle with simple reasoning and planning in physical environments, such as understanding object permanence or planning household activities. The limitation arises from the fact that LMs are trained only on written text and miss essential embodied knowledge and skills. In this paper, we propose a new paradigm of enhancing LMs by finetuning them with world models, to gain diverse embodied knowledge while retaining their general language capabilities. Our approach deploys an embodied agent in a world model, particularly a simulator of the physical world (VirtualHome), and acquires a diverse set of embodied experiences through both goal-oriented planning and random exploration. These experiences are then used to finetune LMs to teach diverse abilities of reasoning and acting in the physical world, e.g., planning and completing goals, object permanence and tracking, etc. Moreover, it is desirable to preserve the generality of LMs during finetuning, which facilitates generalizing the embodied knowledge across tasks rather than being tied to specific simulations. We thus further introduce the classical (EWC) for selective weight updates, combined with low-rank adapters (LoRA) for training efficiency. Extensive experiments show our approach substantially improves base LMs on 18 downstream tasks by 64.28% on average. In particular, the small LMs (1.3B, 6B, and 13B) enhanced by our approach match or even outperform much larger LMs (e.g., ChatGPT).

cs.CL↗

arXiv2023

Agree to Disagree: Measuring Hidden Dissent in FOMC Meetings

Using FOMC transcripts and customized deep learning models, we quantify ``hidden dissent'', or disagreement in the FOMC that is unobserved in formal votes. We find hidden dissent to be prevalent and systematically driven by macroeconomic conditions like inflation and unemployment. It strongly correlates with divergent member projections (SEP) and measures of policy sub-optimality, reflecting heterogeneity among members in policy preferences. Furthermore, we show that the financial markets respond to the hidden dissent implied in FOMC minutes.

econ.GN↗

arXiv2023

FlexMoE: Scaling Large-scale Sparse Pre-trained Model Training via Dynamic Device Placement

With the increasing data volume, there is a trend of using large-scale pre-trained models to store the knowledge into an enormous number of model parameters. The training of these models is composed of lots of dense algebras, requiring a huge amount of hardware resources. Recently, sparsely-gated Mixture-of-Experts (MoEs) are becoming more popular and have demonstrated impressive pretraining scalability in various downstream tasks. However, such a sparse conditional computation may not be effective as expected in practical systems due to the routing imbalance and fluctuation problems. Generally, MoEs are becoming a new data analytics paradigm in the data life cycle and suffering from unique challenges at scales, complexities, and granularities never before possible. In this paper, we propose a novel DNN training framework, FlexMoE, which systematically and transparently address the inefficiency caused by dynamic dataflow. We first present an empirical analysis on the problems and opportunities of training MoE models, which motivates us to overcome the routing imbalance and fluctuation problems by a dynamic expert management and device placement mechanism. Then we introduce a novel scheduling module over the existing DNN runtime to monitor the data flow, make the scheduling plans, and dynamically adjust the model-to-hardware mapping guided by the real-time data traffic. A simple but efficient heuristic algorithm is exploited to dynamically optimize the device placement during training. We have conducted experiments on both NLP models (e.g., BERT and GPT) and vision models (e.g., Swin). And results show FlexMoE can achieve superior performance compared with existing systems on real-world workloads -- FlexMoE outperforms DeepSpeed by 1.70x on average and up to 2.10x, and outperforms FasterMoE by 1.30x on average and up to 1.45x.

cs.DC↗

arXiv2022

TreeMix: Compositional Constituency-based Data Augmentation for Natural Language Understanding

Data augmentation is an effective approach to tackle over-fitting. Many previous works have proposed different data augmentations strategies for NLP, such as noise injection, word replacement, back-translation etc. Though effective, they missed one important characteristic of language--compositionality, meaning of a complex expression is built from its sub-parts. Motivated by this, we propose a compositional data augmentation approach for natural language understanding called TreeMix. Specifically, TreeMix leverages constituency parsing tree to decompose sentences into constituent sub-structures and the Mixup data augmentation technique to recombine them to generate new sentences. Compared with previous approaches, TreeMix introduces greater diversity to the samples generated and encourages models to learn compositionality of NLP data. Extensive experiments on text classification and SCAN demonstrate that TreeMix outperforms current state-of-the-art data augmentation methods.

cs.CL↗

arXiv2022

Don't Take It Literally: An Edit-Invariant Sequence Loss for Text Generation

Neural text generation models are typically trained by maximizing log-likelihood with the sequence cross entropy (CE) loss, which encourages an exact token-by-token match between a target sequence with a generated sequence. Such training objective is sub-optimal when the target sequence is not perfect, e.g., when the target sequence is corrupted with noises, or when only weak sequence supervision is available. To address the challenge, we propose a novel Edit-Invariant Sequence Loss (EISL), which computes the matching loss of a target n-gram with all n-grams in the generated sequence. EISL is designed to be robust to various noises and edits in the target sequences. Moreover, the EISL computation is essentially an approximate convolution operation with target n-grams as kernels, which is easy to implement and efficient to compute with existing libraries. To demonstrate the effectiveness of EISL, we conduct experiments on a wide range of tasks, including machine translation with noisy target sequences, unsupervised text style transfer with only weak training signals, and non-autoregressive generation with non-predefined generation order. Experimental results show our method significantly outperforms the common CE loss and other strong baselines on all the tasks. EISL has a simple API that can be used as a drop-in replacement of the CE loss: https://github.com/guangyliu/EISL.

cs.CL↗

arXiv2021

$\mathbfβ$-delayed proton emission from $\mathbf{^{11}}$Be in effective field theory

We calculate the rate of the rare decay $^{11}\text{Be}$ into $^{10}\text{Be} + p +e^- + \barν_e$ using Halo effective field theory, thereby describing the process of beta-delayed proton emission. We assume a shallow $1/2^+$ resonance in the $^{10}\text{Be}-p$ system with an energy consistent with a recent experiment by Ayyad et al. and obtain $b_p = 4.9_{-2.9}^{+5.6}\text{(exp.)}_{-0.8}^{+4.0}\text{(theo.)} \times 10^{-6}$ for the branching ratio of this decay, predicting a resonance width of $Γ_R = (9.0^{+4.8}_{-3.3}\text{(exp.)}^{+5.3}_{-2.2}\text{(theo.)})~\text{keV}$. Our calculation shows that the experimental branching ratio and resonance parameters of Ayyad et al. are consistent with each other. Moreover, we analyze the general impact of a resonance on the branching ratio and demonstrate that a wide range of combinations of resonance energies and widths can reproduce branching ratios of the correct order. Thus, no exotic mechanism (such as beyond the standard model physics) is needed to explain the experimental decay rate.

nucl-th↗

arXiv2021

Progressive Generation of Long Text with Pretrained Language Models

Large-scale language models (LMs) pretrained on massive corpora of text, such as GPT-2, are powerful open-domain text generators. However, as our systematic examination reveals, it is still challenging for such models to generate coherent long passages of text (e.g., 1000 tokens), especially when the models are fine-tuned to the target domain on a small corpus. Previous planning-then-generation methods also fall short of producing such long text in various domains. To overcome the limitations, we propose a simple but effective method of generating text in a progressive manner, inspired by generating images from low to high resolution. Our method first produces domain-specific content keywords and then progressively refines them into complete passages in multiple stages. The simple design allows our approach to take advantage of pretrained LMs at each stage and effectively adapt to any target domain given only a small set of examples. We conduct a comprehensive empirical study with a broad set of evaluation metrics, and show that our approach significantly improves upon the fine-tuned large LMs and various planning-then-generation methods in terms of quality and sample efficiency. Human evaluation also validates that our model generations are more coherent.

cs.CL↗

arXiv2020

Electric dipole moments of three-nucleon systems in the pionless effective field theory

We calculate the electric dipole moments (EDMs) of three-nucleon systems at leading order in pionless effective field theory. The one-body contributions that arise from permanent proton and neutron EDMs and the two-body contributions that arise from CP-odd nucleon-nucleon interactions are taken into account. Neglecting the Coulomb interaction, we consider the triton and ${}^3$He, and also investigate them in the Wigner-SU(4) symmetric limit. We also calculate the electric dipole form factor and find numerically that the momentum dependence of the electric dipole form factor in the Wigner limit is, up to an overall constant (and numerical accuracy), the same as the momentum dependence of the charge form factor.

nucl-th↗

arXiv2020

Data-to-Text Generation with Style Imitation

Recent neural approaches to data-to-text generation have mostly focused on improving content fidelity while lacking explicit control over writing styles (e.g., word choices, sentence structures). More traditional systems use templates to determine the realization of text. Yet manual or automatic construction of high-quality templates is difficult, and a template acting as hard constraints could harm content fidelity when it does not match the record perfectly. We study a new way of stylistic control by using existing sentences as soft templates. That is, the model learns to imitate the writing style of any given exemplar sentence, with automatic adaptions to faithfully describe the content record. The problem is challenging due to the lack of parallel data. We develop a neural approach that includes a hybrid attention-copy mechanism, learns with weak supervisions, and is enhanced with a new content coverage constraint. We conduct experiments in restaurants and sports domains. Results show our approach achieves stronger performance than a range of comparison methods. Our approach balances well between content fidelity and style control given exemplars that match the records to varying degrees.

cs.CL↗

arXiv2020

Local Additivity Based Data Augmentation for Semi-supervised NER

Named Entity Recognition (NER) is one of the first stages in deep language understanding yet current NER models heavily rely on human-annotated data. In this work, to alleviate the dependence on labeled data, we propose a Local Additivity based Data Augmentation (LADA) method for semi-supervised NER, in which we create virtual samples by interpolating sequences close to each other. Our approach has two variations: Intra-LADA and Inter-LADA, where Intra-LADA performs interpolations among tokens within one sentence, and Inter-LADA samples different sentences to interpolate. Through linear additions between sampled training data, LADA creates an infinite amount of labeled data and improves both entity and context learning. We further extend LADA to the semi-supervised setting by designing a novel consistency loss for unlabeled data. Experiments conducted on two NER benchmarks demonstrate the effectiveness of our methods over several strong baselines. We have publicly released our code at https://github.com/GT-SALT/LADA.

cs.CL↗

arXiv2020

MixText: Linguistically-Informed Interpolation of Hidden Space for Semi-Supervised Text Classification

This paper presents MixText, a semi-supervised learning method for text classification, which uses our newly designed data augmentation method called TMix. TMix creates a large amount of augmented training samples by interpolating text in hidden space. Moreover, we leverage recent advances in data augmentation to guess low-entropy labels for unlabeled data, hence making them as easy to use as labeled data.By mixing labeled, unlabeled and augmented data, MixText significantly outperformed current pre-trained and fined-tuned models and other state-of-the-art semi-supervised learning methods on several text classification benchmarks. The improvement is especially prominent when supervision is extremely limited. We have publicly released our code at https://github.com/GT-SALT/MixText.

cs.CL↗