arXiv subjects

Tianqi Liu

Publications and source records attributed to Tianqi Liu.

63 records · Page 4Linked to original sources

arXiv2022

All Birds with One Stone: Multi-task Text Classification for Efficient Inference with One Forward Pass

Multi-Task Learning (MTL) models have shown their robustness, effectiveness, and efficiency for transferring learned knowledge across tasks. In real industrial applications such as web content classification, multiple classification tasks are predicted from the same input text such as a web article. However, at the serving time, the existing multitask transformer models such as prompt or adaptor based approaches need to conduct N forward passes for N tasks with O(N) computation cost. To tackle this problem, we propose a scalable method that can achieve stronger performance with close to O(1) computation cost via only one forward pass. To illustrate real application usage, we release a multitask dataset on news topic and style classification. Our experiments show that our proposed method outperforms strong baselines on both the GLUE benchmark and our news dataset. Our code and dataset are publicly available at https://bit.ly/mtop-code.

cs.CL↗

arXiv2022

Training ELECTRA Augmented with Multi-word Selection

Pre-trained text encoders such as BERT and its variants have recently achieved state-of-the-art performances on many NLP tasks. While being effective, these pre-training methods typically demand massive computation resources. To accelerate pre-training, ELECTRA trains a discriminator that predicts whether each input token is replaced by a generator. However, this new task, as a binary classification, is less semantically informative. In this study, we present a new text encoder pre-training method that improves ELECTRA based on multi-task learning. Specifically, we train the discriminator to simultaneously detect replaced tokens and select original tokens from candidate sets. We further develop two techniques to effectively combine all pre-training tasks: (1) using attention-based networks for task-specific heads, and (2) sharing bottom layers of the generator and the discriminator. Extensive experiments on GLUE and SQuAD datasets demonstrate both the effectiveness and the efficiency of our proposed method.

cs.CL↗

arXiv2021

NewsEmbed: Modeling News through Pre-trained Document Representations

Effectively modeling text-rich fresh content such as news articles at document-level is a challenging problem. To ensure a content-based model generalize well to a broad range of applications, it is critical to have a training dataset that is large beyond the scale of human labels while achieving desired quality. In this work, we address those two challenges by proposing a novel approach to mine semantically-relevant fresh documents, and their topic labels, with little human supervision. Meanwhile, we design a multitask model called NewsEmbed that alternatively trains a contrastive learning with a multi-label classification to derive a universal document encoder. We show that the proposed approach can provide billions of high quality organic training examples and can be naturally extended to multilingual setting where texts in different languages are encoded in the same semantic space. We experimentally demonstrate NewsEmbed's competitive performance across multiple natural language understanding tasks, both supervised and unsupervised.

cs.CL↗

arXiv2020

Online monitoring for safe pedestrian-vehicle interactions

As autonomous systems begin to operate amongst humans, methods for safe interaction must be investigated. We consider an example of a small autonomous vehicle in a pedestrian zone that must safely maneuver around people in a free-form fashion. We investigate two key questions: How can we effectively integrate pedestrian intent estimation into our autonomous stack. Can we develop an online monitoring framework to give formal guarantees on the safety of such human-robot interactions. We present a pedestrian intent estimation framework that can accurately predict future pedestrian trajectories given multiple possible goal locations. We integrate this into a reachability-based online monitoring scheme that formally assesses the safety of these interactions with nearly real-time performance (approximately 0.3 seconds). These techniques are integrated on a test vehicle with a complete in-house autonomous stack, demonstrating effective and safe interaction in real-world experiments.

cs.RO↗

arXiv2019

A Deep Probabilistic Model for Customer Lifetime Value Prediction

Accurate predictions of customers' future lifetime value (LTV) given their attributes and past purchase behavior enables a more customer-centric marketing strategy. Marketers can segment customers into various buckets based on the predicted LTV and, in turn, customize marketing messages or advertising copies to serve customers in different segments better. Furthermore, LTV predictions can directly inform marketing budget allocations and improve real-time targeting and bidding of ad impressions. One challenge of LTV modeling is that some customers never come back, and the distribution of LTV can be heavy-tailed. The commonly used mean squared error (MSE) loss does not accommodate the significant fraction of zero value LTV from one-time purchasers and can be sensitive to extremely large LTV's from top spenders. In this article, we model the distribution of LTV given associated features as a mixture of zero point mass and lognormal distribution, which we refer to as the zero-inflated lognormal (ZILN) distribution. This modeling approach allows us to capture the churn probability and account for the heavy-tailedness nature of LTV at the same time. It also yields straightforward uncertainty quantification of the point prediction. The ZILN loss can be used in both linear models and deep neural networks (DNN). For model evaluation, we recommend the normalized Gini coefficient to quantify model discrimination and decile charts to assess model calibration. Empirically, we demonstrate the predictive performance of our proposed model on two real-world public datasets.

stat.AP↗

arXiv2019

BERT for Large-scale Video Segment Classification with Test-time Augmentation

This paper presents our approach to the third YouTube-8M video understanding competition that challenges par-ticipants to localize video-level labels at scale to the pre-cise time in the video where the label actually occurs. Ourmodel is an ensemble of frame-level models such as GatedNetVLAD and NeXtVLAD and various BERT models withtest-time augmentation. We explore multiple ways to ag-gregate BERT outputs as video representation and variousways to combine visual and audio information. We proposetest-time augmentation as shifting video frames to one leftor right unit, which adds variety to the predictions and em-pirically shows improvement in evaluation metrics. We firstpre-train the model on the 4M training video-level data, andthen fine-tune the model on 237K annotated video segment-level data. We achieve MAP@100K 0.7871 on private test-ing video segment data, which is ranked 9th over 283 teams.

cs.CV↗

arXiv2018

Constrained-size Tensorflow Models for YouTube-8M Video Understanding Challenge

This paper presents our 7th place solution to the second YouTube-8M video understanding competition which challenges participates to build a constrained-size model to classify millions of YouTube videos into thousands of classes. Our final model consists of four single models aggregated into one tensorflow graph. For each single model, we use the same network architecture as in the winning solution of the first YouTube-8M video understanding competition, namely Gated NetVLAD. We train the single models separately in tensorflow's default float32 precision, then replace weights with float16 precision and ensemble them in the evaluation and inference stages., achieving 48.5% compression rate without loss of precision. Our best model achieved 88.324% GAP on private leaderboard. The code is publicly available at https://github.com/boliu61/youtube-8m

cs.CV↗

arXiv2017

Characterizing Spatiotemporal Transcriptome of Human Brain via Low Rank Tensor Decomposition

Spatiotemporal gene expression data of the human brain offer insights on the spa- tial and temporal patterns of gene regulation during brain development. Most existing methods for analyzing these data consider spatial and temporal profiles separately with the implicit assumption that different brain regions develop in similar trajectories, and that the spatial patterns of gene expression remain similar at different time points. Al- though these analyses may help delineate gene regulation either spatially or temporally, they are not able to characterize heterogeneity in temporal dynamics across different brain regions, or the evolution of spatial patterns of gene regulation over time. In this article, we develop a statistical method based on low rank tensor decomposition to more effectively analyze spatiotemporal gene expression data. We generalize the clas- sical principal component analysis (PCA) which is applicable only to data matrices, to tensor PCA that can simultaneously capture spatial and temporal effects. We also propose an efficient algorithm that combines tensor unfolding and power iteration to estimate the tensor principal components, and provide guarantees on their statistical performances. Numerical experiments are presented to further demonstrate the mer- its of the proposed method. An application of our method to a spatiotemporal brain expression data provides insights on gene regulation patterns in the brain.

stat.ME↗

arXiv2016

Ultrahigh Dimensional Feature Selection via Kernel Canonical Correlation Analysis

High-dimensional variable selection is an important issue in many scientific fields, such as genomics. In this paper, we develop a sure independence feature screening pro- cedure based on kernel canonical correlation analysis (KCCA-SIS, for short). KCCA- SIS is easy to be implemented and applied. Compared to the sure independence screen- ing procedure based on the Pearson correlation (SIS, for short) developed by Fan and Lv [2008], KCCA-SIS can handle nonlinear dependencies among variables. Compared to the sure independence screening procedure based on the distance correlation (DC- SIS, for short) proposed by Li et al. [2012], KCCA-SIS is scale free, distribution free and has better approximation results based on the universal characteristic of Gaussian Kernel (Micchelli et al. [2006]). KCCA-SIS is more general than SIS and DC-SIS in the sense that SIS and DC-SIS correspond to certain choice of kernels. Compared to supremum of Hilbert Schmidt independence criterion-Sure independence screening (sup-HSIC-SIS, for short) developed by Balasubramanian et al. [2013], KCCA-SIS is scale free removing the marginal variation of features and response variables. No model assumption is needed between response and predictors to apply KCCA-SIS and it can be used in ultrahigh dimensional data analysis. Similar to DC-SIS and sup- HSIC-SIS, KCCA-SIS can also be used directly to screen grouped predictors and for multivariate response variables. We show that KCCA-SIS has the sure screening prop- erty, and has better performance through simulation studies. We applied KCCA-SIS to study Autism genes in a spatiotemporal gene expression dataset for human brain development, and obtained better results based on gene ontology enrichment analysis comparing to the other existing methods.

stat.ME↗