arXiv subjects

Xian Li

Publications and source records attributed to Xian Li.

At least 109 records · Page 6Linked to original sources

arXiv2022

Efficient Large Scale Language Modeling with Mixtures of Experts

Mixture of Experts layers (MoEs) enable efficient scaling of language models through conditional computation. This paper presents a detailed empirical study of how autoregressive MoE language models scale in comparison with dense models in a wide range of settings: in- and out-of-domain language modeling, zero- and few-shot priming, and full-shot fine-tuning. With the exception of fine-tuning, we find MoEs to be substantially more compute efficient. At more modest training budgets, MoEs can match the performance of dense models using $\sim$4 times less compute. This gap narrows at scale, but our largest MoE model (1.1T parameters) consistently outperforms a compute-equivalent dense model (6.7B parameters). Overall, this performance gap varies greatly across tasks and domains, suggesting that MoE and dense models generalize differently in ways that are worthy of future study. We make our code and models publicly available for research use.

cs.CL↗

arXiv2022

ATST: Audio Representation Learning with Teacher-Student Transformer

Self-supervised learning (SSL) learns knowledge from a large amount of unlabeled data, and then transfers the knowledge to a specific problem with a limited number of labeled data. SSL has achieved promising results in various domains. This work addresses the problem of segment-level general audio SSL, and proposes a new transformer-based teacher-student SSL model, named ATST. A transformer encoder is developed on a recently emerged teacher-student baseline scheme, which largely improves the modeling capability of pre-training. In addition, a new strategy for positive pair creation is designed to fully leverage the capability of transformer. Extensive experiments have been conducted, and the proposed model achieves the new state-of-the-art results on almost all of the downstream tasks.

eess.AS↗

arXiv2022

Clinicopathological correlation of p40/TTF1 co-expression in NSCLC and review of related literature

TTF1 and ΔNp63/p40 have been used to differentiate ADC and SQC in hypofractionated NSCLC because of their sensitivity and specificity. There are few cases where TTF1 and ΔNp63/p40 are expressed together in the same tumour cells, and little is known about the clinicopathological features, treatment and prognosis of such cases. We investigated the electron microscopic features, immunohistochemical expression and molecular variation of a case of TTF1/p40 co-expressing NSCLC and reviewed and summarised the relevant literature. Our patient was a 58-year-old male with a CT showing a left-sided lung occupancy. As in all other cases reported in the literature, the tumour showed a solid growth pattern with polygonal cells, eosinophilic cytoplasm and clearly visible nuclear fission. Immunohistochemistry showed positive for TTF-1, p40, CK5/6, CK7, P63 and p53, and negative for NapsinA and ALK. Electron microscopy showed tumour cells characterised by bidirectional differentiation of adenocytes and squamous cells, consistent with previous reports. Second-generation sequencing suggested co-mutation of STK11/LKB1 and NF1 genes in this case. Mutations in STK11/LKB1 and NF1 genes have been found in ADC and SQC and are often associated with drug resistance and poor prognosis, but STK11/NF1 co-mutation has not been reported and more cases are needed to reveal the association. p40/TTF1 co-expression in NSCLC may be an under-recognised variant of NSCLC The origin may be a double positive stem cell-like basal cell located in the distal airway, with rapid clinical progression and poor prognosis.

q-bio.TO↗

arXiv2022

An Insight-HXMT view of the mHz quasi-regular modulation phenomenon in the black hole X-ray binary 4U 1630-47

Here we report the spectral-timing results of the black hole X-ray binary 4U 1630-47 during its 2021 outburst using observations from the Hard X-ray Modulation Telescope. Type-C quasi-periodic oscillations (QPOs) in 1.6--4.2 Hz and quasi-regular modulation (QRM) near 60 mHz are detected during the outburst. The mHz QRM has a fractional rms of 10%--16% in the 8--35 keV energy band with a Q factor (frequency/width) of 2--4. Benefiting from the broad energy band of hxmt, we study the energy dependence of the 60 mHz QRM in 1--100 keV for the first time. We find that the fractional rms of the mHz QRM increases with photon energy, while the time lags of the mHz QRM are soft and decrease with photon energy. Fast recurrence of the mHz QRM, in a timescale of less than one hour, has been observed during the outburst. During this period, the corresponding energy spectra moderately change when the source transitions from the QRM state to the non-QRM state. The QRM phenomena also shows a dependence with the accretion rate. We suggest that the QRM could be caused by an unknown accretion instability aroused from the corona.

astro-ph.HE↗

arXiv2022

OPT: Open Pre-trained Transformer Language Models

Large language models, which are often trained for hundreds of thousands of compute days, have shown remarkable capabilities for zero- and few-shot learning. Given their computational cost, these models are difficult to replicate without significant capital. For the few that are available through APIs, no access is granted to the full model weights, making them difficult to study. We present Open Pre-trained Transformers (OPT), a suite of decoder-only pre-trained transformers ranging from 125M to 175B parameters, which we aim to fully and responsibly share with interested researchers. We show that OPT-175B is comparable to GPT-3, while requiring only 1/7th the carbon footprint to develop. We are also releasing our logbook detailing the infrastructure challenges we faced, along with code for experimenting with all of the released models.

cs.CL↗

arXiv2022

Recipes for Adapting Pre-trained Monolingual and Multilingual Models to Machine Translation

There has been recent success in pre-training on monolingual data and fine-tuning on Machine Translation (MT), but it remains unclear how to best leverage a pre-trained model for a given MT task. This paper investigates the benefits and drawbacks of freezing parameters, and adding new ones, when fine-tuning a pre-trained model on MT. We focus on 1) Fine-tuning a model trained only on English monolingual data, BART. 2) Fine-tuning a model trained on monolingual data from 25 languages, mBART. For BART we get the best performance by freezing most of the model parameters, and adding extra positional embeddings. For mBART we match or outperform the performance of naive fine-tuning for most language pairs with the encoder, and most of the decoder, frozen. The encoder-decoder attention parameters are most important to fine-tune. When constraining ourselves to an out-of-domain training set for Vietnamese to English we see the largest improvements over the fine-tuning baseline.

cs.CL↗

arXiv2022

Multilingual Neural Machine Translation with Deep Encoder and Multiple Shallow Decoders

Recent work in multilingual translation advances translation quality surpassing bilingual baselines using deep transformer models with increased capacity. However, the extra latency and memory costs introduced by this approach may make it unacceptable for efficiency-constrained applications. It has recently been shown for bilingual translation that using a deep encoder and shallow decoder (DESD) can reduce inference latency while maintaining translation quality, so we study similar speed-accuracy trade-offs for multilingual translation. We find that for many-to-one translation we can indeed increase decoder speed without sacrificing quality using this approach, but for one-to-many translation, shallow decoders cause a clear quality drop. To ameliorate this drop, we propose a deep encoder with multiple shallow decoders (DEMSD) where each shallow decoder is responsible for a disjoint subset of target languages. Specifically, the DEMSD model with 2-layer decoders is able to obtain a 1.8x speedup on average compared to a standard transformer model with no drop in translation quality.

cs.CL↗

arXiv2022

Efficient Language Modeling with Sparse all-MLP

All-MLP architectures have attracted increasing interest as an alternative to attention-based models. In NLP, recent work like gMLP shows that all-MLPs can match Transformers in language modeling, but still lag behind in downstream tasks. In this work, we analyze the limitations of MLPs in expressiveness, and propose sparsely activated MLPs with mixture-of-experts (MoEs) in both feature and input (token) dimensions. Such sparse all-MLPs significantly increase model capacity and expressiveness while keeping the compute constant. We address critical challenges in incorporating conditional computation with two routing strategies. The proposed sparse all-MLP improves language modeling perplexity and obtains up to 2$\times$ improvement in training efficiency compared to both Transformer-based MoEs (GShard, Switch Transformer, Base Layers and HASH Layers) as well as dense Transformers and all-MLPs. Finally, we evaluate its zero-shot in-context learning performance on six downstream tasks, and find that it surpasses Transformer-based MoEs and dense Transformers.

cs.CL↗

arXiv2022

SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition

Deep face recognition has achieved great success due to large-scale training databases and rapidly developing loss functions. The existing algorithms devote to realizing an ideal idea: minimizing the intra-class distance and maximizing the inter-class distance. However, they may neglect that there are also low quality training images which should not be optimized in this strict way. Considering the imperfection of training databases, we propose that intra-class and inter-class objectives can be optimized in a moderate way to mitigate overfitting problem, and further propose a novel loss function, named sigmoid-constrained hypersphere loss (SFace). Specifically, SFace imposes intra-class and inter-class constraints on a hypersphere manifold, which are controlled by two sigmoid gradient re-scale functions respectively. The sigmoid curves precisely re-scale the intra-class and inter-class gradients so that training samples can be optimized to some degree. Therefore, SFace can make a better balance between decreasing the intra-class distances for clean examples and preventing overfitting to the label noise, and contributes more robust deep face recognition models. Extensive experiments of models trained on CASIA-WebFace, VGGFace2, and MS-Celeb-1M databases, and evaluated on several face recognition benchmarks, such as LFW, MegaFace and IJB-C databases, have demonstrated the superiority of SFace.

cs.CV↗

arXiv2022

Lifting the Curse of Multilinguality by Pre-training Modular Transformers

Multilingual pre-trained models are known to suffer from the curse of multilinguality, which causes per-language performance to drop as they cover more languages. We address this issue by introducing language-specific modules, which allows us to grow the total capacity of the model, while keeping the total number of trainable parameters per language constant. In contrast with prior work that learns language-specific components post-hoc, we pre-train the modules of our Cross-lingual Modular (X-Mod) models from the start. Our experiments on natural language inference, named entity recognition and question answering show that our approach not only mitigates the negative interference between languages, but also enables positive transfer, resulting in improved monolingual and cross-lingual performance. Furthermore, our approach enables adding languages post-hoc with no measurable drop in performance, no longer limiting the model usage to the set of pre-trained languages.

cs.CL↗

arXiv2022

OA-Mine: Open-World Attribute Mining for E-Commerce Products with Weak Supervision

Automatic extraction of product attributes from their textual descriptions is essential for online shopper experience. One inherent challenge of this task is the emerging nature of e-commerce products -- we see new types of products with their unique set of new attributes constantly. Most prior works on this matter mine new values for a set of known attributes but cannot handle new attributes that arose from constantly changing data. In this work, we study the attribute mining problem in an open-world setting to extract novel attributes and their values. Instead of providing comprehensive training data, the user only needs to provide a few examples for a few known attribute types as weak supervision. We propose a principled framework that first generates attribute value candidates and then groups them into clusters of attributes. The candidate generation step probes a pre-trained language model to extract phrases from product titles. Then, an attribute-aware fine-tuning method optimizes a multitask objective and shapes the language model representation to be attribute-discriminative. Finally, we discover new attributes and values through the self-ensemble of our framework, which handles the open-world challenge. We run extensive experiments on a large distantly annotated development set and a gold standard human-annotated test set that we collected. Our model significantly outperforms strong baselines and can generalize to unseen attributes and product types.

cs.CL↗

arXiv2022

Unified Speech-Text Pre-training for Speech Translation and Recognition

We describe a method to jointly pre-train speech and text in an encoder-decoder modeling framework for speech translation and recognition. The proposed method incorporates four self-supervised and supervised subtasks for cross modality learning. A self-supervised speech subtask leverages unlabelled speech data, and a (self-)supervised text to text subtask makes use of abundant text training data. Two auxiliary supervised speech tasks are included to unify speech and text modeling space. Our contribution lies in integrating linguistic information from the text corpus into the speech pre-training. Detailed analysis reveals learning interference among subtasks. Two pre-training configurations for speech translation and recognition, respectively, are presented to alleviate subtask interference. Our experiments show the proposed method can effectively fuse speech and text information into one model. It achieves between 1.7 and 2.3 BLEU improvement above the state of the art on the MuST-C speech translation dataset and comparable WERs to wav2vec 2.0 on the Librispeech speech recognition task.

cs.CL↗

arXiv2022

Quasi-periodic oscillations of the X-ray burst from the magnetar SGR J1935+2154 and associated with the fast radio burst FRB 200428

The origin(s) and mechanism(s) of fast radio bursts (FRBs), which are short radio pulses from cosmological distances, have remained a major puzzle since their discovery. We report a strong Quasi-Periodic Oscillation(QPO) of 40 Hz in the X-ray burst from the magnetar SGR J1935+2154 and associated with FRB 200428, significantly detected with the Hard X-ray Modulation Telescope (Insight-HXMT) and also hinted by the Konus-Wind data. QPOs from magnetar bursts have only been rarely detected; our 3.4 sigma (p-value is 2.9e-4) detection of the QPO reported here reveals the strongest QPO signal observed from magnetars (except in some very rare giant flares), making this X-ray burst unique among magnetar bursts. The two X-ray spikes coinciding with the two FRB pulses are also among the peaks of the QPO. Our results suggest that at least some FRBs are related to strong oscillation processes of neutron stars. We also show that we may overestimate the significance of the QPO signal and underestimate the errors of QPO parameters if QPO exists only in a fraction of the time series of a X-ray burst which we use to calculate the Leahy-normalized periodogram.

astro-ph.HE↗

arXiv2022

PGE: Robust Product Graph Embedding Learning for Error Detection

Although product graphs (PGs) have gained increasing attentions in recent years for their successful applications in product search and recommendations, the extensive power of PGs can be limited by the inevitable involvement of various kinds of errors. Thus, it is critical to validate the correctness of triples in PGs to improve their reliability. Knowledge graph (KG) embedding methods have strong error detection abilities. Yet, existing KG embedding methods may not be directly applicable to a PG due to its distinct characteristics: (1) PG contains rich textual signals, which necessitates a joint exploration of both text information and graph structure; (2) PG contains a large number of attribute triples, in which attribute values are represented by free texts. Since free texts are too flexible to define entities in KGs, traditional way to map entities to their embeddings using ids is no longer appropriate for attribute value representation; (3) Noisy triples in a PG mislead the embedding learning and significantly hurt the performance of error detection. To address the aforementioned challenges, we propose an end-to-end noise-tolerant embedding learning framework, PGE, to jointly leverage both text information and graph structure in PG to learn embeddings for error detection. Experimental results on real-world product graph demonstrate the effectiveness of the proposed framework comparing with the state-of-the-art approaches.

cs.SI↗

arXiv2022

Adaptive Sparse Transformer for Multilingual Translation

Multilingual machine translation has attracted much attention recently due to its support of knowledge transfer among languages and the low cost of training and deployment compared with numerous bilingual models. A known challenge of multilingual models is the negative language interference. In order to enhance the translation quality, deeper and wider architectures are applied to multilingual modeling for larger model capacity, which suffers from the increased inference cost at the same time. It has been pointed out in recent studies that parameters shared among languages are the cause of interference while they may also enable positive transfer. Based on these insights, we propose an adaptive and sparse architecture for multilingual modeling, and train the model to learn shared and language-specific parameters to improve the positive transfer and mitigate the interference. The sparse architecture only activates a sub-network which preserves inference efficiency, and the adaptive design selects different sub-networks based on the input languages. Our model outperforms strong baselines across multiple benchmarks. On the large-scale OPUS dataset with $100$ languages, we achieve $+2.1$, $+1.3$ and $+6.2$ BLEU improvements in one-to-many, many-to-one and zero-shot tasks respectively compared to standard Transformer without increasing the inference cost.

cs.CL↗

arXiv2021

Robust Optimization for Multilingual Translation with Imbalanced Data

Multilingual models are parameter-efficient and especially effective in improving low-resource languages by leveraging crosslingual transfer. Despite recent advance in massive multilingual translation with ever-growing model and data, how to effectively train multilingual models has not been well understood. In this paper, we show that a common situation in multilingual training, data imbalance among languages, poses optimization tension between high resource and low resource languages where the found multilingual solution is often sub-optimal for low resources. We show that common training method which upsamples low resources can not robustly optimize population loss with risks of either underfitting high resource languages or overfitting low resource ones. Drawing on recent findings on the geometry of loss landscape and its effect on generalization, we propose a principled optimization algorithm, Curvature Aware Task Scaling (CATS), which adaptively rescales gradients from different tasks with a meta objective of guiding multilingual training to low-curvature neighborhoods with uniformly low loss for all languages. We ran experiments on common benchmarks (TED, WMT and OPUS-100) with varying degrees of data imbalance. CATS effectively improved multilingual optimization and as a result demonstrated consistent gains on low resources ($+0.8$ to $+2.2$ BLEU) without hurting high resources. In addition, CATS is robust to overparameterization and large batch size training, making it a promising training method for massive multilingual models that truly improve low resource languages.

cs.CL↗

arXiv2021

Do Language Models Have Beliefs? Methods for Detecting, Updating, and Visualizing Model Beliefs

Do language models have beliefs about the world? Dennett (1995) famously argues that even thermostats have beliefs, on the view that a belief is simply an informational state decoupled from any motivational state. In this paper, we discuss approaches to detecting when models have beliefs about the world, and we improve on methods for updating model beliefs to be more truthful, with a focus on methods based on learned optimizers or hypernetworks. Our main contributions include: (1) new metrics for evaluating belief-updating methods that focus on the logical consistency of beliefs, (2) a training objective for Sequential, Local, and Generalizing model updates (SLAG) that improves the performance of learned optimizers, and (3) the introduction of the belief graph, which is a new form of interface with language models that shows the interdependencies between model beliefs. Our experiments suggest that models possess belief-like qualities to only a limited extent, but update methods can both fix incorrect model beliefs and greatly improve their consistency. Although off-the-shelf optimizers are surprisingly strong belief-updating baselines, our learned optimizers can outperform them in more difficult settings than have been considered in past work. Code is available at https://github.com/peterbhase/SLAG-Belief-Updating

cs.CL↗

arXiv2021

Energy-Efficient Online Data Sensing and Processing in Wireless Powered Edge Computing Systems

This paper focuses on developing energy-efficient online data processing strategy of wireless powered MEC systems under stochastic fading channels. In particular, we consider a hybrid access point (HAP) transmitting RF energy to and processing the sensing data offloaded from multiple WDs. Under an average power constraint of the HAP, we aim to maximize the long-term average data sensing rate of the WDs while maintaining task data queue stability. We formulate the problem as a multi-stage stochastic optimization to control the energy transfer and task data processing in sequential time slots. Without the knowledge of future channel fading, it is very challenging to determine the sequential control actions that are tightly coupled by the battery and data buffer dynamics. To solve the problem, we propose an online algorithm named LEESE that applies the perturbed Lyapunov optimization technique to decompose the multi-stage stochastic problem into per-slot deterministic optimization problems. We show that each per-slot problem can be equivalently transformed into a convex optimization problem. To facilitate online implementation in large-scale MEC systems, instead of solving the per-slot problem with off-the-shelf convex algorithms, we propose a block coordinate descent (BCD)-based method that produces close-to-optimal solution in less than 0.04\% of the computation delay. Simulation results demonstrate that the proposed LEESE algorithm can provide 21.9\% higher data sensing rate than the representative benchmark methods considered, while incurring sub-millisecond computation delay suitable for real-time control under fading channel.

cs.IT↗