System.Xml.XmlElement
May 2021’s Gwern.net newsletter is now out; previous, April 2021 ( archives ).
2021 年 5 月的 Gwern.net 时事通讯现已发布；上一篇，2021 年 4 月（档案）。
This is a collation of links and summary of major changes, overlapping with my Changelog ; brought to you by my donors on Patreon .
这是主要更改的链接和摘要的整理，与我的更改日志重叠；由我的 Patreon 捐助者带给您。
Proposal : “Choose Your Own Adventure AI Dungeon” ; “Decision Transformers: Preference Learning As Simple As Possible”
提案：“选择你自己的冒险AI地下城”； “决策变压器：偏好学习尽可能简单”
Hardware : “Podracer architectures for scalable Reinforcement Learning” , Hessel et al 2021 (highly-efficient TPU pod use: eg solving Pong in “Google details new TPUv4 AI accelerator chips” (2.7× TPUv3 chips; up to TPUv4-4096 pods, yielding >1 ExaFLOPS; public access later in 2021)x
硬件：“用于可扩展强化学习的 Podracer 架构”，Hessel 等人，2021 年（高效 TPU Pod 使用：例如在“Google 详细介绍新的 TPUv4 AI 加速器芯片”中解决 Pong（2.7× TPUv3 芯片；最多 TPUv4-4096 个 Pod，产生 >1 ExaFLOPS；2021 年晚些时候公开访问）x
“ZeRO-Infinity: Breaking the GPU Memory Wall for Extreme Scale Deep Learning” , Rajbhandari et al 2021 (~1 trillion parameters per 16 GPUs/DGX-2-node, scaling to >512 GPUs ~40% efficiency)
“ZeRO-Infinity：打破超大规模深度学习的 GPU 内存墙”，Rajbhandari 等人，2021 年（每 16 个 GPU/DGX-2 节点约 1 万亿个参数，可扩展到 > 512 个 GPU，效率约 40%）
“GSPMD: General and Scalable Parallelization for ML Computation Graphs” , Xu et al 2021 (Google upgrade of GPipe / GShard arch to match MS DeepSpeed : “…50%–62% compute utilization on 128–2048 Cloud TPUv3 cores for models with up to one trillion parameters”)
“GSPMD：ML 计算图的通用且可扩展的并行化”，Xu 等人 2021（Google 升级 GPipe / GShard 架构以匹配 MS DeepSpeed：“…对于参数高达 1 万亿的模型，128-2048 个 Cloud TPUv3 核心的计算利用率为 50%-62%”）
“DLRM: High-performance, Distributed Training of Large-scale Deep Learning Recommendation Models” , Mudigere et al 2021 (ZionEX software/hardware platform for training extremely large embeddings—while embeddings aren’t ‘real’ parameters & things like DynamicEmbedding will never learn tricks like GPT-3 no matter how big, they present similar challenges); “RecPipe: Co-designing Models and Hardware to Jointly Optimize Recommendation Quality and Performance” , Gupta et al 2021
“DLRM：大规模深度学习推荐模型的高性能、分布式训练”，Mudigere 等人 2021（ZionEX 软件/硬件平台，用于训练极大的嵌入，而嵌入不是“真实”参数，像 DynamicEmbedding 这样的东西永远不会学习像 GPT-3 这样的技巧，无论有多大，它们都会带来类似的挑战）； “RecPipe：共同设计模型和硬件，共同优化推荐质量和性能”，Gupta 等人，2021 年
“From Motor Control to Team Play in Simulated Humanoid Football” , Liu et al 2021 (curriculum training of a single NN from raw humanoid control to coordinated team-wide soccer strategy; neat to compare with Hill et al 2020 in terms of agent abilities)
“From Motor Control to Team Play in Simulated Humanoid Football”，Liu et al 2021（单个神经网络的课程训练，从原始人形控制到协调全队足球策略；在智能体能力方面与 Hill 等人 2020 进行比较）
“Anthropic” public-benefit-corp/startup launched (founded by the Amodeis; $124M investment for scaling “reliable and steerable AI systems”); “Cooperative AI Foundation” (CAIF) launched
推出“Anthropic”公益公司/初创公司（由 Amodeis 创立；投资 1.24 亿美元用于扩展“可靠且可操纵的人工智能系统”）； “合作人工智能基金会”（CAIF）启动
“MLP-Mixer: An all-MLP Architecture for Vision” , Tolstikhin et al 2021 (another FC paper removing even more inductive biases—ponies are all you need: “Mixer improves more rapidly with data than ResNets, or even ViT, and the gap between large scale Mixer and ViT models shrinks until the performance is matched on the entire dataset…” The Bitter Lesson truly is the single bitterest lesson in ML, isn’t it?
“MLP-Mixer：视觉的全 MLP 架构”，Tolstikhin 等人 2021（另一篇 FC 论文消除了更多归纳偏差 - 小马就是您所需要的一切：“Mixer 在数据方面的改进比 ResNet，甚至 ViT 更快，并且大规模 Mixer 和 ViT 模型之间的差距缩小，直到性能在整个数据集上匹配......” 痛苦的教训确实是 ML 中最痛苦的教训，不是吗？
The more people tweet about how MLP-Mixer is overhyped because is −X% worse than the ultra-hand-optimized baseline or requires Y× more FLOPS, the more they demonstrate precisely why this sort of research is so important!
越多的人在 Twitter 上谈论 MLP-Mixer 被过度炒作，因为它比超手工优化的基线差 −X%，或者需要 Y× 多的 FLOPS，他们就越能准确地证明为什么此类研究如此重要！
And showing, incidentally, that Transformers are still under-researched if such a fundamental fact could have been missed for so long.)
顺便说一句，如果这么一个基本事实被忽视了这么久，那么变形金刚的研究仍然不足。）
“Data-Efficient Language-Supervised Zero-Shot Learning with Self-Distillation” , Cheng et al 2021 ( CLIP -like performance scaled down to n = 3m using soft labels generated by a Conceptual Captions -pretrained model)
“Data-Efficient Language-Supervised Zero-Shot Learning with Self-Distillation”，Cheng et al 2021（使用概念字幕预训练模型生成的软标签，类似 CLIP 的性能缩小到 n = 3m）
“SR3: Image Super-Resolution via Iterative Refinement” , Saharia et al 2021; “Diffusion Models Beat GANs on Image Synthesis” , Dhariwal & Nichol 2021 ( DDPM ^ 1 ^ finally surpass BigGAN-deep on ImageNet 512px images at similar compute-cost, as expected from their good scaling ); “Cascaded Diffusion Models for High Fidelity Image Generation” , Ho et al 2021
“SR3：通过迭代细化实现图像超分辨率”，Saharia 等人，2021 年； “Diffusion Models Beat GANs on Image Synthesis”，Dhariwal & Nichol 2021（DDPM ^ 1 ^ 最终在 ImageNet 512px 图像上以相似的计算成本超越 BigGAN-deep，正如其良好的缩放所预期的那样）； “用于高保真图像生成的级联扩散模型”，Ho 等人，2021 年
“Grokking: Generalization Beyond Overfitting On Small Algorithmic Data Sets” , Power et al 2021 ( discussion ; new scaling effect, ‘grokking’: sudden perfect generalization emerging many epochs after training-set overfitting on algorithmic tasks when training in flat shallow loss landscapes ); “Knowledge distillation: A good teacher is patient and consistent” , Beyer et al 2021 (training much smaller models merely requires hundreds of thousands or millions of epochs)
“Grokking：在小型算法数据集上超越过度拟合的泛化”，Power et al 2021（讨论；新的缩放效应，‘grokking’：在平坦浅层损失景观中训练时，在算法任务上训练集过度拟合后突然出现完美泛化许多纪元）； “知识蒸馏：一个好的老师是耐心和一致的”，Beyer 等人 2021（训练更小的模型只需要数十万或数百万个 epoch）
“Reward is enough” , Silver et al 2021 (a DRL manifesto: reward losses enough at scale of compute/parameters/tasks to induce all important capabilities like memory/exploration/generalization/imitation/reasoning)
“奖励就足够了”，Silver 等人 2021（DRL 宣言：在计算/参数/任务规模上奖励足够的损失，以诱导记忆/探索/泛化/模仿/推理等所有重要能力）
Scaling Down : lazy : a tool for running processes in idle time (how to train on a GPU without destroying your GUI’s usability!
缩小规模：惰性：一种在空闲时间运行进程的工具（如何在 GPU 上进行训练而不破坏 GUI 的可用性！
lazy pauses runs briefly while you interact with your desktop, letting you do months-long runs without going crazy or resorting to Colab etc. This enables hobbyists to go after previously-infeasible model sizes); EleutherAI releases a 6b-parameter GPT-3 model, GPT-J (are you still using GPT-2/GPT-Neo?
当您与桌面交互时，延迟暂停会短暂运行，让您可以进行长达数月的运行，而无需发疯或诉诸 Colab 等。这使爱好者能够追求以前不可行的模型尺寸）； EleutherAI发布6b参数GPT-3模型GPT-J（你还在用GPT-2/GPT-Neo吗？
upgrade!); “Aggregating Nested Transformers” , Zhang et al 2021/ “Less is More: Pay Less Attention in Vision Transformers” , Pan et al 2021
升级！）； “聚合嵌套 Transformers”，Zhang 等人 2021/“少即是多：在视觉 Transformers 中减少注意力”，Pan 等人 2021
“ByT5: Towards a token-free future with pre-trained byte-to-byte models” , Xue et al 2021 (character models—not just feasible but desirable; we’ll get our rhyming & pun-making language models yet!)
“ByT5：通过预先训练的字节到字节模型迈向无代币的未来”，Xue 等人，2021（字符模型——不仅可行，而且令人向往；我们将获得押韵和双关语语言模型！）
“Machine Learning Attacks Against the Asirra CAPTCHA” , Golle 2008 (a look back on a decade of CV progress: months of work for 80% cat vs dog with SVM ensembles in 2008; 5min in Fast.ai for 99% accuracy in 2018; for even more perspective, Cireşan 2012 )
“针对 Asirra CAPTCHA 的机器学习攻击”，Golle 2008（回顾十年的 CV 进展：2008 年使用 SVM 集成进行了 80% 的猫与狗的数月工作；2018 年 Fast.ai 花费 5 分钟实现了 99% 的准确率；要获得更多视角，Cireşan 2012）
Everything Is Heritable: “Bi-ancestral depression GWAS in the Million Veteran Program and meta-analysis in >1.2 million individuals highlight new therapeutic directions” , Levey et al 2021
一切都是可遗传的：“百万退伍军人计划中的双祖抑郁症 GWAS 和超过 120 万人的荟萃分析突出了新的治疗方向”，Levey 等人 2021
“Long read sequencing of 3,622 Icelanders provides insight into the role of structural variants in human diseases and other traits” , Beyter et al 2021
“对 3,622 名冰岛人进行长读长测序，深入了解结构变异在人类疾病和其他性状中的作用”，Beyter 等人，2021
“Rapid Sequencing–Based Diagnosis of Thiamine Metabolism Dysfunction Syndrome” (sequence everyone!)
“基于快速测序的硫胺素代谢功能障碍综合征诊断”（对所有人进行测序！）
Engineering: “Sense codon reassignment enables viral resistance and encoded polymer synthesis” , Robertson et al 2021 (“ultra-safe cells”: synthesizing an entire E.
工程：“正义密码子重新分配能够实现病毒抗性和编码聚合物合成”，Robertson 等人 2021（“超安全细胞”：合成整个大肠杆菌）
coli genome with swapped codons for complete viral immunity)
具有交换密码子的大肠杆菌基因组以实现完全病毒免疫）
“In vivo CRISPR base editing of PCSK9 durably lowers cholesterol in primates” , Musunuru et al 2021
“PCSK9 的体内 CRISPR 碱基编辑可持久降低灵长类动物的胆固醇”，Musunuru 等人 2021
Optogenetics : “Partial recovery of visual function in a blind patient after optogenetic therapy” , Sahel et al 2021 ( media ); “Wireless multilateral devices for optogenetic studies of individual and social behaviors” , Yang et al 2021 ( media )
光遗传学：“光遗传学治疗后盲人患者视觉功能部分恢复”，Sahel 等人 2021（媒体）； “用于个体和社会行为光遗传学研究的无线多边设备”，Yang 等人 2021（媒体）
“Retron Library Recombineering (RLR): High-throughput functional variant screens via in vivo production of single-stranded DNA” , Schubert et al 2021
“Retron 文库重组工程 (RLR)：通过体内生产单链 DNA 进行高通量功能变异筛选”，Schubert 等人 2021
“Genomic characterization of world’s longest selection experiment in mouse reveals the complexity of polygenic traits” , Palma-Vera et al 2021
“世界上最长的小鼠选择实验的基因组特征揭示了多基因性状的复杂性”，Palma-Vera 等人 2021
“Surrogate broodstock to enhance biotechnology research and applications in aquaculture” , Jin et al 2021
“替代亲鱼加强水产养殖生物技术研究和应用”，Jin 等人，2021
“Utility of polygenic embryo screening for disease depends on the selection strategy” , Lencz et al 2021
“多基因胚胎疾病筛查的效用取决于选择策略”，Lencz 等人 2021
“Limit on lab-grown human embryos dropped by stem-cell body: The International Society for Stem Cell Research relaxed the famous 14-day rule on culturing human embryos in its latest research guidelines”
“实验室培养的人类胚胎被干细胞体丢弃的限制：国际干细胞研究学会在其最新研究指南中放宽了著名的培养人类胚胎的14天规定”
“How a Publicity Blitz Created The Myth of Subliminal Advertising” , Rogers 1992 (the famous movie-theater/popcorn-sales experiment never happened)
“宣传闪电战如何创造了潜意识广告的神话”，罗杰斯 1992 年（著名的电影院/爆米花销售实验从未发生过）
“Clarifying the Structure and Nature of Left-Wing Authoritarianism (LWA)” , Costello et al 2021
“澄清左翼威权主义 (LWA) 的结构和性质”，Costello 等人 2021
“A connectomic study of a petascale fragment of human cerebral cortex” , Shapson-Coe et al 2021 (“…This “digital tissue” is a ~660,000× scale up of an earlier saturated reconstruction from a small region of mouse cortex, published in 2015 ( Kasthuri et al 2015 ).
“人类大脑皮层千万亿级片段的连接组学研究”，Shapson-Coe 等人，2021（“……这种“数字组织”是对小鼠皮层小区域的早期饱和重建的约 660,000 倍放大，发表于 2015 年（Kasthuri 等人，2015 年）。
Although this scaleup was difficult, it was not hundreds of thousands of times more difficult and took about the same amount of time as the previous data set (~4 years)…The rapid improvements over the past few years…argues that analyzing volumes that are even 3 orders of magnitude larger, such as an exascale whole mouse brain connectome, will likely be in reach within a decade." See also “Accelerating progress in brain recording tech” .)
尽管这种扩展很困难，但它的难度并没有增加数十万倍，并且花费的时间与之前的数据集大致相同（约 4 年）……过去几年的快速改进……认为分析甚至大 3 个数量级的体积（例如百亿亿级整个小鼠大脑连接组）可能会在十年内实现。”另请参阅“加速大脑记录技术的进步”。）
“Neuroimaging evidence for a network sampling theory of individual differences in human intelligence test performance” , Soreq et al 2021; “The neural basis of intelligence in fine-grained cortical topographies” , Feilong et al 2021; “Predicting intelligence from brain gray matter volume” , Hilger et al 2020 (towards the mechanistic reification of g : per P-FIT , it is global efficiency/total cognitive resources which can be spent on learning & orchestrating specialized capabilities); if we consider recent human brain imaging studies, cross-species comparisons, and deep learning as converging, I would offer as a speculation the following:
“人类智力测试表现个体差异的网络抽样理论的神经影像证据”，Soreq 等人，2021 年； “细粒度皮质拓扑中智力的神经基础”，Feilong et al 2021； “根据大脑灰质体积预测智力”，Hilger et al 2020（朝着 g 的机械具体化：根据 P-FIT，它是可用于学习和编排专业能力的全局效率/总认知资源）；如果我们认为最近的人类大脑成像研究、跨物种比较和深度学习是趋同的，我会提出以下推测：
The Master Synthesis: intelligence is execution of small simplicity-weighted programs, best discovered by search over smooth loss landscapes like that of highly-overparameterized differentiable networks containing lottery-ticket subnetworks which are ensembled/averaged over, approaching Bayes-optimal reasoning in the limit (as nearest-neighbors-like high dimensional interpolation / memorization gives way to algorithmic generalization / interpolation on a more abstract level); this can be implemented by large numbers of similar neurons trained using any of the many approximations to backprop; human intelligence’s g is real but is the overall ‘pool’ of neural resources which derives from overall body integrity because the number of neurons, their density, their myelination, resistance to damage and infection etc, is causally downstream of all body and developmental systems, creating a huge mutational target; the brain regions specialize and differentiate, and their orchestration (or lack thereof) contributes to observed performance on tasks tapping into multiple specialized regions; as tasks rely on fewer regions or approach intrinsic ceiling, g ceases to be observable and task-specific influences matter most.
主综合：智能是小型简单加权程序的执行，最好通过搜索平滑损失景观来发现，例如高度过度参数化的可微分网络，其中包含彩票子网络，这些子网络被集成/平均，在极限范围内接近贝叶斯最优推理（就像最近邻居一样的高维插值/记忆让位于更抽象级别的算法泛化/插值）；这可以通过使用反向传播的许多近似值中的任何一个来训练大量相似的神经元来实现；人类智力的 g 是真实的，但它是源自整体身体完整性的神经资源的整体“池”，因为神经元的数量、密度、髓鞘形成、对损伤和感染的抵抗力等，是所有身体和发育系统的因果下游，创造了一个巨大的突变目标；大脑区域专门化和分化，它们的编排（或缺乏编排）有助于观察到利用多个专门区域的任务表现；随着任务依赖的区域减少或接近内在上限，g 不再是可观察到的，特定任务的影响最为重要。
“MDMA-assisted therapy for severe PTSD: a randomized, double-blind, placebo-controlled phase 3 study” , Mitchell et al 2021 ( d = 0.9 over therapy); “Effects of Psilocybin-Assisted Therapy on Major Depressive Disorder” , Davis et al 2021
“MDMA 辅助治疗严重 PTSD：一项随机、双盲、安慰剂对照 3 期研究”，Mitchell 等人，2021 年（治疗结束后 d = 0.9）； “裸盖菇素辅助治疗对重度抑郁症的影响”，Davis 等人 2021
“Why Animals Don’t Get Lost: Birds do it.
“为什么动物不会迷路：鸟类会迷路。
Bees do it.
蜜蜂做到了。
Learning about the astounding navigational feats of wild creatures can teach us a lot about where we’re going” (on spectacular but still mysterious feats of animal navigation )
了解野生动物令人惊叹的导航壮举可以教会我们很多关于我们要去哪里的知识”（关于动物导航的壮观但仍然神秘的壮举）
“Linking Brain Biology to Intellectual Endowment: A Review on the Associations of Human Intelligence With Neuroimaging Data” , Dizaji et al 2021
“将大脑生物学与智力禀赋联系起来：人类智力与神经影像数据关联的回顾”，Dizaji 等人 2021
“The Best And The Rest: Revisiting The Norm Of Normality Of Individual Performance” , O’Boyle & Aguinis 2012 (performance is log-normal )
“最好的和其他的：重新审视个人表现正常性的规范”，O’Boyle & Aguinis 2012（表现是对数正态）
“A conserved strategy for inducing appendage regeneration” , Abrams et al 2021 (slight regrowth of damaged mouse limbs by drinking sugar+amino-acid-supplemented water)
“诱导附肢再生的保守策略”，Abrams et al 2021（通过饮用糖+氨基酸补充水使受损小鼠四肢轻微再生）
“Feeling Small: Exploring the Tactile Perception Limits [of Humans]” , Skedung et al 2013
“感觉渺小：探索[人类]的触觉感知极限”，Skedung 等人 2013
“The Board Game of the Alpha Nerds: Before Risk , before Dungeons & Dragons , before Magic: The Gathering , there was Diplomacy ” ( WP ; “I still don’t know whom I should have trusted, if anyone.
“阿尔法书呆子的棋盘游戏：在《冒险》之前，在《龙与地下城》之前，在《万智牌》之前，有《外交》”（WP；“我仍然不知道我应该信任谁，如果有的话。
All I know is that I felt stupid, stressed out, humiliated, and sad.”)
我只知道我感到愚蠢、压力、羞辱和悲伤。”）
“I walk the (beta-stability) line: How counting neutrons explains nuclear waste”
“我走在（β稳定性）路线：计算中子如何解释核废料”
“Spintronics: Build mechanical circuits” , Kickstarter (followup to Turing Tumble )
“自旋电子学：构建机械电路”，Kickstarter（图灵翻滚的后续）
“RCTs to Scale: Comprehensive Evidence from 2 Nudge Units” , DellaVigna & Linos 2020 (nudge effects overestimated by 6.2× due to publication bias)
“随机对照试验规模化：来自 2 个助推单位的综合证据”，DellaVigna 和 Linos 2020（由于发表偏差，助推效果被高估了 6.2 倍）
“No causal associations between childhood family income and subsequent psychiatric disorders, substance misuse and violent crime arrests: a nationwide Finnish study of >650,000 individuals and their siblings” , Sariaslan et al 2021; “Parental income and mental disorders in children and adolescents: prospective register-based study” , Kinge et al 2021
“童年家庭收入与随后的精神疾病、药物滥用和暴力犯罪逮捕之间没有因果关系：一项针对超过 650,000 人及其兄弟姐妹的芬兰全国研究”，Sariaslan 等人，2021 年； “父母收入与儿童和青少年的精神障碍：基于登记的前瞻性研究”，Kinge 等人 2021
“The Strange Story of Dagobert, the Duck Tales Bandit: In the ’90s, a frustrated artist in Berlin went on a crime spree—building bombs, extorting high-end stores, and styling his persona after Scrooge McDuck.
《鸭子故事强盗达戈伯特的奇怪故事》：90 年代，柏林一位失意的艺术家开始了犯罪狂潮——制造炸弹、勒索高端商店，并模仿斯克鲁奇·麦克老鸭 (Scrooge McDuck) 塑造自己的形象。
He soon became a German folk hero.” ( WP ; another reminder for Americans—odd as it may seem, Donald Duck is extremely popular overseas; see also the unknown-in-the-USA character John D.
他很快就成为德国民间英雄。” （ WP ；对美国人的另一个提醒——尽管看起来很奇怪，唐老鸭在海外非常受欢迎；也可以看看在美国不知名的角色约翰·D·唐老鸭（Donald Duck）。
Rockerduck or beloved Scandinavian tradition From All of Us to All of You who 2020 airing set an all-time record of >4.5m viewers)
Rockerduck 或深受喜爱的斯堪的纳维亚传统《From All of Us to All of You》，2020 年播出，创下了超过 450 万观众的历史记录）
List of atmospheric optical phenomena (How many would you recognize from a distance or plane?
大气光学现象列表（从远处或平面上您能认出多少种？
How many have you even heard of?)
你听说过多少个？）
Baron Franz Nopcsa von Felső-Szilvás (noted geologist, paleontologist, anthropologist, homosexual, & skyjacker)
弗兰茨·诺普萨·冯·费尔索-西尔瓦斯男爵（著名地质学家、古生物学家、人类学家、同性恋者和劫机者）
What is a diffusion model like DDPM?
什么是 DDPM 之类的扩散模型？
To try to explain it as simply as possible without the math :
尝试在不使用数学的情况下尽可能简单地解释它：
DDPM is a neural net which is trained to fix noise in an image: it takes a noisy image and ‘sharpens’ it to produce a new image.
DDPM 是一种神经网络，经过训练可以修复图像中的噪声：它获取噪声图像并对其进行“锐化”以生成新图像。
You train it by adding dirt to a normal image, and teaching it to turn the dirty version into the original.
您可以通过向正常图像添加污垢来训练它，并教它将脏版本变成原始图像。
As it gets better, it learns what the images all tend to look like so it can ‘see through’ ever more noise, to turn smudged hints of the original image into its best guess.
随着它变得更好，它会了解所有图像的外观，这样它就可以“看穿”更多的噪音，将原始图像的污迹暗示转变为最佳猜测。
Once it’s done training, what happens if you give it a completely dirty photo, which is pure static noise?
一旦完成训练，如果你给它一张完全脏的照片（纯粹的静态噪声）会发生什么？
Well, it produces a slightly less dirty ‘photo’.
嗯，它会生成一张不太脏的“照片”。
And if you do it again?
如果你再做一次呢？
it’s a little cleaner still.
还是比较干净一点。
Now, what if you do this many times?
现在，如果你多次这样做怎么办？
It has to get cleaner each time.
每次它都必须变得更干净。
The end result: the static noise goes in, and a face pops out!
最终结果：静电噪声进入，一张脸突然出现！
The DDPM has hallucinated a face out of the noise.
DDPM 从噪音中幻出了一张脸。
One little blob of static here turned into a nose, and another blob turned into an ear, and it went from there.
这里的一小团静电变成了鼻子，另一团静电变成了耳朵，然后从那里出去。