<?xml version="1.0" encoding="UTF-8" ?>
<rss version="2.0">
    <channel>
      <title>AI Research Wiki</title>
      <link>https://ruiyizhang.com/wiki/ai-research</link>
      <description>Last 10 notes on AI Research Wiki</description>
      <generator>Quartz -- quartz.jzhao.xyz</generator>
      <item>
    <title>BERT (Bidirectional Encoder Representations from Transformers)</title>
    <link>https://ruiyizhang.com/wiki/ai-research/architectures/bert-architecture</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/architectures/bert-architecture</guid>
    <description><![CDATA[ High-Level Design BERT, introduced by Devlin et al. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Encoder-Decoder Architecture</title>
    <link>https://ruiyizhang.com/wiki/ai-research/architectures/encoder-decoder</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/architectures/encoder-decoder</guid>
    <description><![CDATA[ High-Level Design The encoder-decoder (or seq2seq) architecture processes an input sequence through an encoder network that produces a sequence of hidden representations, then feeds those representations to a decoder network that generates an output sequence autoregressively. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Generative Pre-trained Transformer (GPT)</title>
    <link>https://ruiyizhang.com/wiki/ai-research/architectures/gpt</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/architectures/gpt</guid>
    <description><![CDATA[ High-Level Design The GPT family uses a decoder-only transformer architecture trained autoregressively on next-token prediction. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Mixture of Experts (MoE)</title>
    <link>https://ruiyizhang.com/wiki/ai-research/architectures/mixture-of-experts-architecture</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/architectures/mixture-of-experts-architecture</guid>
    <description><![CDATA[ High-Level Design The Mixture of Experts (MoE) architecture is a sparse model design where each input token is routed to a small subset of specialized “expert” sub-networks rather than passing through the entire model. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Transformer</title>
    <link>https://ruiyizhang.com/wiki/ai-research/architectures/transformer</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/architectures/transformer</guid>
    <description><![CDATA[ High-Level Design The transformer is a sequence-to-sequence architecture introduced by Vaswani et al. in “attention-is-all-you-need” (2017). ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Attention</title>
    <link>https://ruiyizhang.com/wiki/ai-research/concepts/attention</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/concepts/attention</guid>
    <description><![CDATA[ Definition Attention is a mechanism that allows neural networks to dynamically focus on relevant parts of their input when producing each element of the output. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Chain-of-Thought Prompting</title>
    <link>https://ruiyizhang.com/wiki/ai-research/concepts/chain-of-thought</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/concepts/chain-of-thought</guid>
    <description><![CDATA[ Definition Chain-of-thought (CoT) prompting is a technique where the model is encouraged to generate intermediate reasoning steps before producing a final answer. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Constitutional AI</title>
    <link>https://ruiyizhang.com/wiki/ai-research/concepts/constitutional-ai</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/concepts/constitutional-ai</guid>
    <description><![CDATA[ Definition Constitutional AI (CAI) is an alignment approach developed by anthropic that trains AI systems to be helpful and harmless using a set of written principles (a “constitution”) rather than relying entirely on human labels for harmlessness. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Direct Preference Optimization</title>
    <link>https://ruiyizhang.com/wiki/ai-research/concepts/direct-preference-optimization</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/concepts/direct-preference-optimization</guid>
    <description><![CDATA[ Definition Direct Preference Optimization (DPO) is an alignment method that optimizes a language model directly from human preference data without training a separate reward model or using reinforcement learning. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item><item>
    <title>Fine-Tuning</title>
    <link>https://ruiyizhang.com/wiki/ai-research/concepts/fine-tuning</link>
    <guid>https://ruiyizhang.com/wiki/ai-research/concepts/fine-tuning</guid>
    <description><![CDATA[ Definition Fine-tuning is the process of adapting a pretrained model to a specific task or behavior by continuing training on task-specific data, typically with a smaller learning rate. ]]></description>
    <pubDate>Sat, 11 Apr 2026 00:00:00 GMT</pubDate>
  </item>
    </channel>
  </rss>