![[Olewave's Review] CLIP (3/3): Learning Transferable Visual Models From Natural Language Supervision](/media/video/2023/01/olewaves-review-clip-3-3-learning-transferable-visual-models-from-natural-language/preview.jpg)
[Olewave评论] CLIP (3/3): 从自然语言监督学习可转移的视觉模型
CLIP评论的最后一部分聚焦于结果部分，这是OpenAI对比图文模型从有趣想法转变为整个多模态堆栈基础原语的时刻。
![[Olewave's Review] CLIP (2/3): Learning Transferable Visual Models From Natural Language Supervision](/media/video/2023/01/olewaves-review-clip-2-3-learning-transferable-visual-models-from-natural-language/preview.jpg)
[Olewave评论] CLIP (2/3): 从自然语言监督学习可转移的视觉模型
OpenAI的CLIP通过抛弃固定标签集，转而在从互联网上抓取的4亿个(图像、文本)对上进行训练，改变了计算机视觉的游戏规则。
![[Olewave's Review] CLIP (1/3): Learning Transferable Visual Models From Natural Language Supervision](/media/video/2022/12/olewaves-review-clip-1-3-learning-transferable-visual-models-from-natural-language/preview.jpg)
![[Olewave's Long Review] Xception: Deep Learning with Depthwise Separable Convolutions](/media/video/2022/07/olewaves-long-review-xception-deep-learning-with-depthwise-separable-convolutions/preview.jpg)
[Olewave长评] Xception: 具有深度可分卷积的深度学习
Xception将Inception假设推向了逻辑极端,通过将跨通道和空间相关性分解为完全独立的操作,而它推广的深度可分卷积现在随处可见
iOS 16 全新听写功能如何工作？一位语音研究员揭秘苹果的秘制配方！
苹果 iOS 16 的听写功能完全在设备本地运行，这一设计选择对延迟、隐私以及手机上可行的 ASR 架构类型产生连锁影响。
![[Long Review] Conformer: Convolution-augmented Transformer for Speech Recognition](/media/video/2022/06/long-review-conformer-convolution-augmented-transformer-for-speech-recognition/preview.jpg)
[长评论] Conformer: 用于语音识别的卷积增强 Transformer
Conformer 是那个悄悄接管端到端 ASR 的模型，其配方——在每个 Transformer 块上添加卷积模块——被证明是那些看似简单但行之有效的想法之一。
![[Long Review] Cascaded Diffusion Models for High Fidelity Image Generation](/media/video/2022/06/long-review-cascaded-diffusion-models-for-high-fidelity-image-generation/preview.jpg)
[长评] 级联扩散模型高保真图像生成
在Imagen和DALL-E 2将扩散模型作为生成图像的默认选择之前，这篇Google Brain论文阐述了许多后来的文本到图像系统悄悄借鉴的级联扩散方案：从小
![[Short Review] Cascaded Diffusion Models for High Fidelity Image Generation](/media/video/2022/06/short-review-cascaded-diffusion-models-for-high-fidelity-image-generation/preview.jpg)
![[Long Review] Axial Attention in Multidimensional Transformers](/media/video/2022/05/long-review-axial-attention-in-multidimensional-transformers/preview.jpg)
[长评] 多维Transformer中的Axial Attention
对图像或视频进行完整自注意力的计算成本太高，那么你如何在不产生二次计算爆炸的情况下保持Transformer的表达能力呢？
![[Short Review] Axial Attention in Multidimensional Transformers](/media/video/2022/05/short-review-axial-attention-in-multidimensional-transformers/preview.jpg)
[短评] 多维Transformer中的Axial Attention
Axial Attention是使Transformer在高维数据上保持可处理性的优雅想法之一：你不是一次对所有像素或所有时间-频率箱进行注意力计算，而是一次沿一个轴进行。
![[Long Review] Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis](/media/video/2022/05/long-review-transfer-learning-from-speaker-verification-to-multispeaker-text-to-speech/preview.jpg)
[长评] 从说话人验证迁移学习到多说话人Text-To-Speech合成
这是使零样本声音克隆实用化的论文：在具有数千个嘈杂、无转录的声音上训练说话人编码器以进行判别验证任务，然后将这些嵌入馈送到Tacotron 2 s
![[Short Review] Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis](/media/video/2022/05/short-review-transfer-learning-from-speaker-verification-to-multispeaker-text-to-speech/preview.jpg)
[短评] 从说话人验证迁移学习到多说话人Text-To-Speech合成
启动现代零样本声音克隆浪潮的Google论文具有美观简洁的架构：基于验证训练的说话人编码器、将文本加说话人嵌入转化的Tacotron 2合成器
![[Short Review] Wav2Seq: Pre-training Speech-to-Text Encoder-Decoder Models Using Pseudo Languages](/media/video/2022/05/short-review-wav2seq-pre-training-speech-to-text-encoder-decoder-models-using-pseudo/preview.jpg)
[短评] Wav2Seq：使用伪语言预训练语音到文本编码器-解码器模型
像wav2vec 2.0这样的自监督预训练给了我们很好的语音编码器，但如果您想要一个完整的编码器-解码器ASR系统，解码器仍然从零开始。
![[Long Review] Towards Zero-Label Language Learning](/media/video/2022/05/long-review-towards-zero-label-language-learning/preview.jpg)
![[Long Review] Fully Sharded Data Parallel: faster AI training with fewer GPUs](/media/video/2022/04/long-review-fully-sharded-data-parallel-faster-ai-training-with-fewer-gpus/preview.jpg)
![[Short Review] Fully Sharded Data Parallel: faster AI training with fewer GPUs](/media/video/2022/04/short-review-fully-sharded-data-parallel-faster-ai-training-with-fewer-gpus/preview.jpg)
[短评] 完全分片数据并行：用更少的GPU进行更快的AI训练
完全分片数据并行是Meta对每个语音和语言团队最终都会提出的问题的答案：我们如何能够训练大一个数量级的模型，而无需配置大一个数量级的GPU？
![[Short Review] Deduplicating Training Data Makes Language Models Better](/media/video/2022/04/short-review-deduplicating-training-data-makes-language-models-better/preview.jpg)