Search

After more than 70 years of evolution, great achievements have been made in machine translation. Especially in recent years, translation quality has been greatly improved with the emergence of neural machine translation (NMT). In this article, we first review the history of machine translation from rule-based machine translation to example-based machine translation and statistical machine translation. We then introduce NMT in more detail, including the basic framework and the current dominant framework, Transformer, as well as multilingual translation models to deal with the data sparseness problem. In addition, we introduce cutting-edge simultaneous translation methods that achieve a balance between translation quality and latency. We then describe various products and applications of machine translation. At the end of this article, we briefly discuss challenges and future research directions in this field.

We report the largest experience, to our knowledge, of home cardiorespiratory monitoring in 47,495 newborns using the novel Owlet Smart Sock (OSS) technology (October 2015 to May 2017). On average, 47,495 newborns were monitored for 6 months, 4.5 d/wk, 9.9 h/d. Continuous readings of oxygen saturation and heart rate were obtained from 39,626 full-term newborns. OSS users were likely first-time parents in their 30s with at least a college degree; 37% had a healthcare professional in the family; and 28% were at or below median income level per the US Census Bureau. “Peace of mind” was the reason to own an OSS in 75%, and 82% of parents followed Safe Sleep Guidelines. A total of 94% of parents reported a better quality of sleep. The fast and continuous pace of device adoption and reported experience suggest excellent parental acceptance of the OSS. Prospective studies are warranted to further evaluate its applications in the high-risk newborn population.

Adolescent idiopathic scoliosis (AIS) is a clinically significant disorder with high heritability that affects 2-4% of the population. Genome-wide association studies have identified LBX1 as a strong susceptibility locus for AIS in Asian and Caucasian populations. Here we further dissect the genetic association with AIS in a Caucasian population. To identify genetic markers associated with AIS we employed a genome-wide association study (GWAS) design comparing 620 female Caucasian patients who developed idiopathic scoliosis during adolescence with 1,287 ethnically matched females who had normal spinal curves by skeletal maturity. The genomic region around LBX1 was imputed and haplotypes investigated for genetic signals under different inheritance models. The strongest signal was identified upstream of LBX1 (rs11190878, P(trend) = 4.18 × 10(-9), OR = 0.63[0.54-0.74]). None of the remaining SNPs pass the genome-wide significance threshold. We found rs11190870, downstream of LBX1 and previously associated with AIS in Asian populations, to be in modest linkage disequilibrium (LD) with rs11190878 (r(2) = 0.40, D' = 0.81). Haplotype analysis shows that rs11190870 and rs11190878 track a single risk factor that resides on the ancestral haplotype and is shared across ethnic groups. We identify six haplotypes at the LBX1 locus including two strongly associated haplotypes; a recessive risk haplotype (TTA, Control(freq) = 0.52, P = 1.25 × 10(-9), OR = 1.56), and a co-dominant protective haplotype (CCG, Control(freq) = 0.28, P = 2.75 × 10(-7), OR = 0.65). Together the association signals from LBX1 explain 1.4% of phenotypic variance. Our results identify two clinically relevant haplotypes in the LBX1-region with opposite effects on AIS risk. The study demonstrates the utility of haplotypes over un-phased SNPs for individualized risk assessment by more strongly delineating individuals at risk for AIS without compromising the effect size.

Endometriosis is a complex gynecological condition that affects 6-10% of women in their reproductive years and is defined by the presence of endometrial glands and stroma outside the uterus. Twin, family, and genome-wide association (GWA) studies have confirmed a genetic role, yet only a small part of the genetic risk can be explained by SNP variation. Copy number variants (CNVs) account for a greater portion of human genetic variation than SNPs and include more recent mutations of large effect. CNVs, likely to be prominent in conditions with decreased reproductive fitness, have not previously been examined as a genetic contributor to endometriosis. Here we employ a high-density genotyping microarray in a genome-wide survey of CNVs in a case-control population that includes 2,126 surgically confirmed endometriosis cases and 17,974 population controls of European ancestry. We apply stringent quality filters to reduce the false positive rate common to many CNV-detection algorithms from 77.7% to 7.3% without noticeable reduction in the true positive rate. We detected no differences in the CNV landscape between cases and controls on the global level which showed an average of 1.92 CNVs per individual with an average size of 142.3 kb. On the local level we identify 22 CNV-regions at the nominal significance threshold (P

Endometriosis is a common gynecological condition with complex etiology defined by the presence of endometrial glands and stroma outside the womb. Endometriosis is a common cause of both cyclic and chronic pelvic pain, reduced fertility, and reduced quality-of-life. Diagnosis and treatment of endometriosis is, on average, delayed by 7-10 years from the onset of symptoms. Absence of a timely and non-invasive diagnostic tool is presently the greatest barrier to the identification and treatment of endometriosis. Twin and family studies have documented an increased relative risk in families. To identify genetic factors that contribute to endometriosis we conducted a two-stage genome-wide association study (GWAS) of a European cohort including 2,019 surgically confirmed endometriosis cases and 14,471 controls. Three of the SNPs we identify associated at P

We suggest that foundation models are general purpose solutions similar to general purpose programmable microprocessors, where fine-tuning and prompt-engineering are analogous to coding for microprocessors. Evaluating general purpose solutions is not like hypothesis testing. We want to know how well the machine will perform on an unknown program with unknown inputs for unknown users with unknown budgets and unknown utility functions. This paper is based on an invited talk by John Mashey, "Lessons from SPEC," at an ACL-2021 workshop on benchmarking. Mashey started by describing Standard Performance Evaluation Corporation (SPEC), a benchmark that has had more impact than benchmarks in our field because SPEC addresses an import commercial question: which CPU should I buy? In addition, SPEC can be interpreted to show that CPUs are 50,000 faster than they were 40 years ago. It is remarkable that we can make such statements without specifying the program, users, task, dataset, etc. It would be desirable to make quantitative statements about improvements of general purpose foundation models over years/decades without specifying tasks, datasets, use cases, etc. [ABSTRACT FROM AUTHOR]

Retrieval-augmented generation (RAG) adds a simple but powerful feature to chatbots, the ability to upload files just-in-time. Chatbots are trained on large quantities of public data. The ability to upload files just-in-time makes it possible to reduce hallucinations by filling in gaps in the knowledge base that go beyond the public training data such as private data and recent events. For example, in a customer service scenario, with RAG, we can upload your private bill and then the bot can discuss questions about your bill as opposed to generic FAQ questions about bills in general. This tutorial will show how to upload files and generate responses to prompts; see https://github.com/kwchurch/RAG for multiple solutions based on tools from OpenAI, LangChain, HuggingFace transformers and VecML. [ABSTRACT FROM AUTHOR]