<<返回 岭南师范学院—学生翻译实践成果展
以下内容为岭南师范学院学生最近一个月内的翻译实践成果

Awesome PyTorch Lightning template TLDR: A PyTorch Lightning template with a lot of features included. Link to the Google Colab here. Epistemic status: This template is the result of few weeks of learning, not years of experience. Think of this as your friends’ lecture notes, not the teachers’ handouts. To show you how under-qualified and over-opinionated I am, just check the list of issues I didn’t managed to solve. And if you know how to solve them, please please please tell me. Have you ever been confused about what is the best practice for PyTorch? Like, when to send your Tensor to GPU? or when to call zero_grad? Or have you tried to do something new, like adding a SimCLR-like pipeline, and having to rewrite a majority of your code because it was so poorly written? Or maybe, you are wondering what everyone else’s pipeline is like? Are theirs 100% more efficient? What are the simple tips and tricks that you have been missing? Maybe not you, but I have. PyTorch Lightning (PL) comes to the rescue. It is basically a template on how your code should be structured. PL has a lot of features in their documentations, like: logging inspecting gradient profiler etc. They also have a lot templates such as: The simplest example called the Boring model for debugging Scratch model for rapid prototyping Basic examples like MNIST Advanced example like Generative Adversarial Network (GAN) And even more stuff in PL Bolts (github) The issue is that all of these example are minimalistic. Which is great for a demo or as a template. But what’s missing is a more complete example showing how all of these features integrate. Especially for people whose first coding project in deep learning, so here it is: The template Link to the Google Colab here. Features Logging Logging to TensorBoard AND stdout for redundancy (I made this myself ^^). I still can’t copy everything. Some information are only put to the stdout. This is an issue because if you use supercomputing clusters, the output files have nondescript names, making it harder to look up experimental details, especially far into the future. (I’m considering logging to WandB as well). Proper use hp_metric so we can select the best hyperparameters within TensorBoard (not working yet T_T. As a temporary work around, it saves to a .json that gets loaded into a pandas DataFrame). Loss curve (I made this myself ^^). Entire script timing (for estimating WALLTIME if you are using supercomputing clusters) (I made this myself ^^). Inspect gradient norms to prevent vanishing or exploding gradient. Log parameters as histogram to TensorBoard (I made this myself ^^). Logging individual parameters might not be realistic, and there would be millions of parameters per epoch. Print a summary of your LightningModule. This is an example of printed output that I cannot redirect to TensorBoard text. Log system (hardware and software) info. Debugging Profiler (PyTorch) to figure out which layers / operations are the bottleneck that have been stealing your time and memory. Note that this slows things down. By a lot! So make sure you turn this off before you go to hyperparameter tuning. Sanity check is a feature that is turned on by default. Good to know that this exist. (I found out about this the hard way) There are 2 ways to monitor GPU. The first one just monitors the memory, while the second one can monitor a number of statistics. My template just uses the first one. There are two ways to run on shorten epochs. The first is by limiting the number of batches. While the second one is fast_dev_run which limits the number of batches in the background anyway, among other things. My template calls the limiting arguments directly. Make model overfit on subset of data (Bug: Profiler clashes with shorten epoch. ) Optimization Early stopping because let’s not waste resources when the model already converged. Gradient Clipping When it comes to optimizer, I used to just simply use Adam, with ReduceLROnPlateau and call it a day (I don’t even optimize for betas). But this stops me from sleeping at night because I always second guess myself, wondering if I’m missing on huge improvements. And I know that this is a VERY active area of research. But the alternative is the curse of dimensionality with optimizer hyperparameters. This is where PL comes to my rescue. Now, I could simply consider PL as an industry standard, use all the optimization tools provided, and sleep a little easier. And here are the two tools: Learning Rate Finder, and Stochastic Weight Averaging. Saving and loading weights Save the best model and tests it. Once you finished your hyperparameter search, you might want to load a checkpoint after the session is closed to do things like residual analysis or activation analysis (look at the pretty figure above) on the best model. (Tracking how the activations changes during training might also be helpful, but I don’t implement this in this template.) Hyperparameter tuning: There is an example of grid search in my Google Colab template. Yes, I know random search is better, but this is just for demo. For some reason, I can’t get the hp_metric on TensorBoard to work, so I made a .json work around. Also included is a snippet to aggregate the .json files from different experimental runs. What I also need is a hyperparameter optimizer library that implements good algorithm, that read and writes from an offline file (because I’m using HPC). The best solution so far is Optuna because it is easy to parallelize offline. Where to go from here? Obviously my template on Google Colab if you haven’t check it. But you might want to check out existing models in Bolts and make your own LightningDataModule for your own datasets. Good luck! References All images, except noted otherwise in the caption, are mine.

2025-11-15 黄静茹 人工智能 英-中

After the start, the international fleet raced on a short sharp beat to the East Bramble mark. First to round was Théa Khelif/Thomas Andre (FRA), followed by Megan Thomson/Oakley Marsh (NZL 2) then Onur Tok/Hamide Sevda Ersezer (TUR). The front runners double-headed to the spreader mark, Paul Heys before hoisting spinnakers. Heading into the Western Solent, FRA 1 showed good speed on the VMG run, extending their lead. Both NZL 2 and TUR chose to follow the mainland shore with the Kiwis notably sailing goose-winged to hold their lane. Alec Coleman/Savannah Taylor (USA), Liz Wardley/Lincoln Dews (AUS 2) and Alec Coleman/Savannah Taylor (CAN) chose the island shore in the Western Solent and looked to make gains. First to round the Needles Fairway Buoy was FRA 1,closely followed by NZL 2 and AUS 2. Only the top five finishers from this qualifier will earn a direct place in the World Championship Final, making every move critical. Live Race Tracking – Follow Every Move All boats are fitted with YB Trackers so fans can follow the fleet in real time across the Solent and English Channel. Watch for free at https://yb.tl/odhwc2025 or via the YB Races App, available on app stores. The Royal Ocean Racing Club (RORC), in collaboration with Cap-Regatta and supported by LGL and Jeanneau, is proud to host the 2025 Offshore Double Handed World Championships from 22 September to 1 October from Cowes, Isle of Wight. Text and images courtesy of RORC.org. Read more here.

2025-11-15 简晓琪 体育 英-中

One in six laboratory-confirmed bacterial infections causing common infections in people worldwide in 2023 were resistant to antibiotic treatments, according to a new World Health Organization (WHO) report launched today. Between 2018 and 2023, antibiotic resistance rose in over 40% of the pathogen-antibiotic combinations monitored, with an average annual increase of 5–15%. Data reported to the WHO Global Antimicrobial Resistance and Use Surveillance System (GLASS) from over 100 countries cautions that increasing resistance to essential antibiotics poses a growing threat to global health. The new Global antibiotic resistance surveillance report 2025 presents, for the first time, resistance prevalence estimates across 22 antibiotics used to treat infections of the urinary and gastrointestinal tracts, the bloodstream and those used to treat gonorrhoea. The report covers 8 common bacterial pathogens – Acinetobacter spp., Escherichia coli, Klebsiella pneumoniae, Neisseria gonorrhoeae, non-typhoidal Salmonella spp., Shigella spp., Staphylococcus aureus and Streptococcus pneumoniae – each linked to one or more of these infections. WHO estimates that antibiotic resistance is highest in the WHO South-East Asian and Eastern Mediterranean Regions, where 1 in 3 reported infections were resistant. In the African Region, 1 in 5 infections was resistant. Resistance is also more common and worsening in places where health systems lack capacity to diagnose or treat bacterial pathogens. “Antimicrobial resistance is outpacing advances in modern medicine, threatening the health of families worldwide,” said Dr Tedros Adhanom Ghebreyesus, WHO Director-General. “As countries strengthen their AMR surveillance systems, we must use antibiotics responsibly, and make sure everyone has access to the right medicines, quality-assured diagnostics, and vaccines. Our future also depends on strengthening systems to prevent, diagnose and treat infections and on innovating with next-generation antibiotics and rapid point-of-care molecular tests.” The new report notes that drug-resistant Gram-negative bacteria are becoming more dangerous worldwide, with the greatest burden falling on countries least equipped to respond. Among these, E. coli and K. pneumoniae are the leading drug-resistant Gram-negative bacteria found in bloodstream infections. These are among the most severe bacterial infections that often result in sepsis, organ failure, and death. Yet more than 40% of E. coli and over 55% of K. pneumoniae globally are now resistant to third-generation cephalosporins, the first-choice treatment for these infections. In the African Region, resistance even exceeds 70%. Other essential life-saving antibiotics, including carbapenems and fluoroquinolones, are losing effectiveness against E. coli, K. pneumoniae, Salmonella, and Acinetobacter. Carbapenem resistance, once rare, is becoming more frequent, narrowing treatment options and forcing reliance on last-resort antibiotics. And such antibiotics are costly, difficult to access, and often unavailable in low- and middle-income countries. Country participation in GLASS has increased over four-fold, from 25 countries in 2016 to 104 countries in 2023. However, 48% of countries did not report data to GLASS in 2023 and about half of the reporting countries still lacked the systems to generate reliable data. In fact, countries facing the largest challenges lacked the surveillance capacity to assess their antimicrobial resistance (AMR) situation. The political declaration on AMR adopted at the United Nations General Assembly in 2024 set targets to address AMR through strengthening health systems and working with a ‘One Health’ approach coordinating across human health, animal health, and environmental sectors. To combat the growing challenge of AMR, countries must commit to strengthening laboratory systems and generating reliable surveillance data, especially from underserved areas, to inform treatments and policies. WHO calls on all countries to report high-quality data on AMR and antimicrobial use to GLASS by 2030. Achieving this target will require concerted action to strengthen the quality, geographic coverage, and sharing of AMR surveillance data to track progress. Countries should scale up coordinated interventions designed to address antimicrobial resistance across all levels of healthcare and ensure that treatment guidelines and essential medicines lists align with local resistance patterns. The report is accompanied by expanded digital content available in the WHO’s GLASS dashboard, which provides global and regional summaries, country profiles based on unadjusted surveillance coverage and AMR data, and detailed information on antimicrobial use.

2025-11-14 温柔 医学 英-中

The Australian company Gilmour Space aims to make a second attempt to reach space in 2026, having turned a cow paddock into a launch pad. Gilmour Space launched its first Eris rocket on July 29 from the Bowen Orbital Spaceport in coastal Queensland, but the rocket fell to Earth just 14 seconds after liftoff. Despite the setback, the TestFlight1 mission laid the foundation for sovereign launch from Australian soil, according to Adam Gilmour, CEO and co-founder of Gilmour Space Technologies, speaking at the International Astronautical Congress (IAC) in Sydney, Australia, on Oct. 3. "We had 14 seconds of flight time, 23 seconds of engine burn time. And we have obviously gotten a lot of data out of it, a lot of information, and we were pretty happy with it," Gilmour said. He added that, on average, it takes a rocket company an average of three launch attempts to get to orbit successfully. Behind July's brief flight lay years of groundwork. The journey to the pad involved challenges and hurdles in terms of engineering, financing, regulations and licensing — including 24 different permits from Queensland, along with environmental and airspace approvals — as well as finding a location for a launch pad. "We drove 2,000 kilometers [1,240 miles] out into the middle of Australia, into an area that I think is very similar to what the surface of Mars looks like," Gilmour said. "When we did the launch, a lot of the media said Gilmour launched a rocket in a cow paddock. I got offended — but then you go down there, and you do see cows." Now, with infrastructure in place, and having worked with regulators, the company is plotting its return to flight for 2026. "We are going to be launching again next year," Gilmour said. "We're going to be doing more launch attempts, so we're not going to give up." This, he added, is the start of what could be a bright future for the Australian space sector. "I think the future looks fantastic … If you look at all the building blocks of what the country's done, we now have rules and regulations that permit orbital launches. "I think there's at least four or five companies in Australia that have satellites operationally working in space, which is really, really good," Gilmour said. "And we are well capitalized. So I think in the next five years, you're going to see more launches out of Australia."

2025-11-14 陈思宇 航空 英-中

三井化学株式会社(简称“三井化学”)与大赛璐株式会社(简称“大赛璐”)的全资子公司宝理塑料株式会社(简称“宝理塑料”)宣布,双方就三井化学开发和销售的工程塑料产品(ARLEN®和AURUM®品牌)的市场营销业务签订了合作协议。 根据该协议,自2026年1月1日起,三井化学将委托宝理塑料负责这些产品的部分市场营销业务。宝理塑料将全权负责客户对接、新客户开发和技术支持等市场营销活动。 三井化学将继续负责ARLEN®和AURUM®的研发、生产、物流等业务。目前暂无调整销售渠道及相关流程的计划。 三井化学旨在通过借助宝理塑料的客户网络与解决方案能力,满足汽车、电子电气领域对高性能工程塑料日益增长的需求,从而实现业务的进一步增长。 同时,宝理塑料通过新增ARLEN®与AURUM®的市场营销业务,将进一步丰富其产品组合,增强其为更多客户提供解决方案的能力。 两家公司将在未来探索进一步深化合作的机会。 三井化学的ARLEN®是一种改性聚酰胺6T,具有高耐热性、低吸水性、高机械强度和良好的流动性,适用于连接器,插口,按钮等SMT电子部件、汽车部件、汽车电装部件、汽缸头部盖板、恒温器外壳等发动机室内部件。 三井化学的AURUM®是一种热塑性聚酰亚胺(TPI),在热塑性树脂中拥有最高级别的耐热性,颠覆了聚酰亚胺的性能卓越加工难的认知,是一款可以通过注塑成型方式实现高效率生产的超级工程塑料。AURUM®适用于多种应用环境,如挖掘机导向环,汽车变速箱和涡轮增压的密封件,有油和无油环境下的各种轴套应用。 2025年10月16日,大赛璐宣布“宝理塑料将根据大赛璐集团的重组计划进行拆分”,并表示大赛璐目前正在考虑自2026年4月1日起承接宝理塑料的工程塑料业务。该合作协议计划自2026年4月1日起转让给大赛璐。

2025-11-13 汪嘉琪 化工 中-英

Dive Brief: Five years after COVID-19 shut down classrooms and shifted college admissions testing policies, the SAT and ACT are still drawing fewer students than during pre-pandemic years. Some 1.38 million students took the ACT in 2025 compared to 1.78 million in 2019, and about 2 million students took the SAT this year versus 2.22 million in 2019, data released recently by the testing companies show. SAT scores, meanwhile, increased only slightly from the high school class of 2024 to the class of 2025, while ACT scores stayed about level. In both cases, scores fell below those from the pre-pandemic year of 2019. Dive Insight: The slight uptick in SAT scores and level ACT scores for the high school graduating class of 2025 are still positive trends compared to last year, when average scores on both tests declined year-over-year compared to 2023. Still, SAT scores were still “substantially lower than average scores prior to the pandemic," said College Board, the organization that publishes the test. In 2025, average SAT scores were 521 in reading and writing and 508 in math. In 2019, those averages were 10 points higher for reading and writing (531) and 20 points higher for math (528). The ACT average composite score, 19.4, also fell lower than the 2019 score of 20.7. For ACT test-takers, 30% met three or more of the four college readiness benchmarks in English, math, reading and science. The ACT benchmarks indicate that students have a 50% chance of earning a B or better in first-year college courses of the same subject and a 75% chance of a C or better. Meanwhile, the dip in overall test takers for both exams continues a trend that dates to at least the pandemic, when colleges shifted toward test-optional policies. For the ACT, however, the numbers began declining much earlier. While testing experts had expected the pandemic to trigger a shift away from K-12 standardized tests, ​​that didn't materialize to a great degree and standardized and high-stakes testing are still core to K-12. More than 90% of four-year colleges in the U.S. were not expected to require applicants for fall 2026 admission to submit ACT or SAT scores, according to data released in September by FairTest, a nonprofit that advocates for limiting college entrance exams. That's over 2,000 of the nation's bachelor-degree granting institutions. Since fall 2020, the number of test-optional or test-free colleges have increased overall, the organization's annual count shows. In the meantime, FairTest said the number of institutions requiring entrance exams minimally increased — from 154 for fall 2025 admissions to 160 for fall 2026 admissions. “While a handful of schools have reinstated testing requirements over the past two admissions cycles for a variety of institutional reasons and in response to external pressures, ACT/SAT-optional and test-blind/score-free policies remain the normal baseline in undergraduate admissions,” said FairTest Executive Director Harry Feder in a September statement. “Test-optional policies continue to dominate at national universities, state flagships, and selective liberal arts colleges.”

2025-11-12 陈思宇 教育资讯 英-中

本报百色10月26日电 (廖景芝 秦塬棋)近日,中国铁路南宁局集团有限公司主动对接地方文旅项目,围绕“2025百色酸啾啾音乐节”等大型活动精准施策,通过“高铁+音乐节”联动模式拓展客运营销,助力旅游淡季客流上量。 本次音乐节规模达2.4万人,作为广西旅游发展大会系列活动的亮点之一,该音乐节带动南宁、柳州等地旅客前往百色。南宁局集团公司提前着手,联合主办方推出“高铁票免费送”专属福利——购买音乐节内场VIP双人套票的观众,可免费获得南宁至百色或返程高铁二等座车票一张,以“高铁+文旅”融合策略激发跨区域出行需求。 为应对音乐节客流,南宁局集团公司多措并举深化淡季营销。他们开展靶向走访,巩固与当地重点旅行社的合作,推介电子发票便捷服务及“旅团”账户优势,拓展团体票源;优化售票服务,为旅客设计联程票、优化中转方案,提升购票体验;精准对接文旅活动,提前掌握今年广西文化旅游发展大会等活动情况,提前部署运力资源。 在服务保障方面,铁路部门同步推出职工专属福利,通过直通购票二维码享受音乐节门票优惠。此外,南宁局集团公司细化票务兑换流程,明确观众凭高铁购票凭证至音乐节现场返现等活动规则,确保活动衔接顺畅。 南宁局集团公司百色车务段业务科科长丁明表示,此次合作是铁路响应“客运提质”计划、深耕“铁路+文旅”模式的重要实践。下一步,该段将继续对接地方文旅节庆、赛事活动,动态优化运输方案,通过“一日一图”、增开列车等方式,实现运力与需求高效匹配,为淡季增运增收提供支撑。 近年来,南宁局集团公司持续探索“铁路+”融合路径,结合区域特色打造“高铁+音乐节”“高铁+研学”等品牌项目,进一步释放客运潜力,推动地方经济社会高质量发展。

2025-11-12 温柔 交通运输 中-英

A white dwarf star has been caught in the middle of consuming a planetary relic, offering fresh clues as to what happens to planetary systems after their star dies. Three billion years ago, a sun-like star reached the end of its life, throwing off its outer layers following its red giant phase to leave behind its inert core, which we see today as the white dwarf called LSPM J0207+3331, located 145 light-years away. But what happened to its planets? Spectroscopic observations of LSPM J0207+3331 by several telescopes, including the Magellan Baade 6.5-meter telescope in Chile and the 10-meter Keck I telescope on Hawaii's Mauna Kea, have revealed that fragments of planets and asteroids have survived for three billion years. For one of those fragments, however, its time is at an end. The spectroscopic observations have shown that gravitational tidal forces from the white dwarf have torn it apart, scattering debris from the planetary body onto the surface of the white dwarf. The measurements identified 13 elements from this doomed object, including aluminum, carbon, chromium, cobalt, copper, iron, magnesium, manganese, nickel, silicon, sodium, strontium and titanium, in mostly Earth-like abundances. The white dwarf features a hydrogen-rich envelope, and in general any elements deposited onto the white dwarf should sink into this hydrogen envelope and disappear from view. For so many elements to still be visible implies that their accretion onto the white dwarf must have happened relatively recently — within the past 35,000 years. It could even be ongoing – LSPM J0207+3331 could still be dismantling this object, which is estimated to have been 120 miles (193 kilometers) across, a piece at a time as you read this. Heavy elements from destroyed planets and asteroids have been detected on white dwarfs before, but after three billion years this process of debris falling onto the white dwarf should have ended. "The amount of rocky material is unusually high for a white dwarf of this age," said Patrick Dufour of the Trottier Institute for Research on Exoplanets at Université de Montréal in a statement. LSPM J0207+3331 is also ringed by a probable debris disk rich in silicates and which was discovered as an excess mid-infrared glow by NASA's Wide-field Infrared Survey Explorer (WISE). The hypothesis is that the object that has recently been ripped apart by the white dwarf could have originated from this debris disk of material that survived the death of the star. Future James Webb Space Telescope (JWST) observations of this disk could allow researchers to determine its mineralogy and constrain its total mass, which will provide further clues as to the nature of the object that the white dwarf has destroyed. What is not yet clear is why has this object met its doom now, and not at any time in the previous three billion years? "This discovery challenges our understanding of planetary system evolution," added Érika Le Bourdais, also of Montréal and the lead author of the research. "Ongoing accretion at this stage suggests white dwarfs may also retain planetary remnants still undergoing dynamical changes." When a sun-like star begins to die and expands into a red giant, its inner planets are consumed and destroyed. However, bodies orbiting far enough away, such as asteroids, comets and gas giant planets, can survive, after a fashion. The changing gravitational field as a star sheds mass can disrupt the planets' orbits, resulting in many collisions between asteroids, comets and surviving planets and moons over billions of years that can grind solid bodies down to dust and small chunks. It's this material that fills the debris disk around LSPM J0207+3331, and what is surprising is that substantial solid bodies still exist in that disk, and that something must have happened to cause one of those solid bodies to fall towards the white dwarf. "Something clearly disturbed this system long after the star's death," said John Debes of the Space Telescope Science Institute. "There's still a reservoir of material capable of polluting the white dwarf, even after billions of years." What has destabilized the debris is unclear. Any surviving gas giant planets could be responsible, with multi-planet interactions having possibly gradually destabilized orbits of smaller bodies over billions of years. "This could point to long-term dynamical processes we don’t yet fully understand," said Debes. Proving this idea won't be easy, however. Gas giant planets would be too far from the white dwarf and likely too cool to be bright enough to be imaged, although the JWST might be able to have a go. More likely, the European Space Agency's Gaia astrometric mission may have been able to detect a wobble in the motion of the white dwarf on the sky caused by the gravity of orbiting gas giant planets pulling on it. The first batch of exoplanet data from Gaia is expected to be released in December 2026 – perhaps then the mystery might be solved? The findings have been published on Oct. 22 in The Astrophysical Journal.

2025-11-12 宁晓敏 航空 英-中

本校翻译实践排行榜

汪嘉琪
8171字
戴晓欣
7763字
冯梓晴
7504字
4
徐苗媚
7362字
5
张艺贤
7282字

400所高校都在用的翻译教学平台

试译宝所属母公司