🐦 Twitter Post Details

Viewing enriched Twitter post

@_akhaliq

Google Deepmind presents Mixtures of Experts Unlock Parameter Scaling for Deep RL paper page: https://t.co/IjxzP9rrV6 The recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.

Media 1
View on Twitter

📊 Media Metadata

{
  "media": [
    {
      "url": "https://crmoxkoizveukayfjuyo.supabase.co/storage/v1/object/public/media/posts/1757605047989379420/media_0.png",
      "type": "photo",
      "original_url": "https://pbs.twimg.com/media/GGREePWWEAApuBT.png",
      "download_date": "2025-08-13T05:51:37.326095",
      "stored_in_supabase": true,
      "format_converted_from_list": true
    }
  ],
  "conversion_date": "2025-08-13T00:32:26.083828",
  "format_converted": true,
  "original_structure": "had_media_only"
}

🔧 Raw API Response

{
  "user": {
    "created_at": "2014-04-27T00:20:12.000Z",
    "default_profile_image": false,
    "description": "AI research paper tweets, ML @Gradio (acq. by @HuggingFace 🤗)\n\ndm for promo",
    "fast_followers_count": 0,
    "favourites_count": 29401,
    "followers_count": 286817,
    "friends_count": 2354,
    "has_custom_timelines": true,
    "is_translator": false,
    "listed_count": 3625,
    "location": "subscribe → ",
    "media_count": 15355,
    "name": "AK",
    "normal_followers_count": 286817,
    "possibly_sensitive": false,
    "profile_banner_url": "https://pbs.twimg.com/profile_banners/2465283662/1610997549",
    "profile_image_url_https": "https://pbs.twimg.com/profile_images/1451191636810092553/kpM5Fe12_normal.jpg",
    "screen_name": "_akhaliq",
    "statuses_count": 26003,
    "translator_type": "none",
    "url": "https://t.co/TbGnXZJwEc",
    "verified": true,
    "withheld_in_countries": [],
    "id_str": "2465283662"
  },
  "id": "1757605047989379420",
  "conversation_id": "1757605047989379420",
  "full_text": "Google Deepmind presents Mixtures of Experts Unlock Parameter Scaling for Deep RL\n\npaper page: https://t.co/IjxzP9rrV6\n\nThe recent rapid progress in (self) supervised learning models is in large part predicted by empirical scaling laws: a model's performance scales proportionally to its size. Analogous scaling laws remain elusive for reinforcement learning domains, however, where increasing the parameter count of a model often hurts its final performance. In this paper, we demonstrate that incorporating Mixture-of-Expert (MoE) modules, and in particular Soft MoEs (Puigcerver et al., 2023), into value-based networks results in more parameter-scalable models, evidenced by substantial performance increases across a variety of training regimes and model sizes. This work thus provides strong empirical evidence towards developing scaling laws for reinforcement learning.",
  "reply_count": 6,
  "retweet_count": 116,
  "favorite_count": 492,
  "hashtags": [],
  "symbols": [],
  "user_mentions": [],
  "urls": [
    {
      "url": "https://t.co/Tp2vpWZrZ8",
      "expanded_url": "https://huggingface.co/papers/2402.08609",
      "display_url": "huggingface.co/papers/2402.08…"
    }
  ],
  "media": [
    {
      "media_url": "https://pbs.twimg.com/media/GGREePWWEAApuBT.png",
      "type": "photo"
    }
  ],
  "url": "https://twitter.com/_akhaliq/status/1757605047989379420",
  "created_at": "2024-02-14T03:17:52.000Z",
  "#sort_index": "1757605047989379420",
  "view_count": 99068,
  "quote_count": 3,
  "is_quote_tweet": false,
  "is_retweet": false,
  "is_pinned": false,
  "is_truncated": true,
  "startUrl": "https://twitter.com/_akhaliq/status/1757605047989379420"
}