{"slug": "arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world", "title": "arXiv Endorsement Request for cs.LG - Diagnostic Control for Hierarchical World Models", "summary": "A researcher preparing a first arXiv submission in cs.LG is seeking an endorsement for a study introducing a random-abstractor control to test hierarchical world models (H-JEPA), finding that only partial observability with a recurrent abstractor yields a real gap (19.91 ± 3.36pp over random, 3 seeds).", "body_md": "Hi everyone,\n\nI’m preparing my first arXiv submission in cs.LG and need an endorsement to submit.\n\nShort summary: H-JEPA (LeCun, 2022) proposes hierarchical joint-embedding prediction but doesn’t specify how to verify a trained abstraction actually encodes anything a random projection of the same shape wouldn’t. I introduce a random-abstractor control (trained vs. untrained abstractor, identical architecture) and run a 2x2 study crossing observability (full/egocentric) with abstractor type (instantaneous/recurrent) in controlled gridworld environments. Three of four conditions produce abstractions statistically indistinguishable from random projections; only partial observability + a recurrent abstractor yields a real, replicated gap (19.91 ± 3.36pp over random, 3 seeds). I also report a negative result on landmark density that didn’t survive multi-seed replication.\n\nIf anyone here is registered as an endorser for cs.LG and willing to take a look, I’d be very grateful. Happy to share the full draft privately.\n\nTo endorse, please visit:\n\nIf that link doesn’t work, visit:\n\nand enter code: MTENXK\n\nThank you!", "url": "https://wpnews.pro/news/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world", "canonical_source": "https://discuss.huggingface.co/t/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world-models/179473#post_1", "published_at": "2026-08-31 05:12:41+00:00", "updated_at": "2026-08-31 05:23:12.035051+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-research"], "entities": ["arXiv", "H-JEPA", "LeCun"], "alternates": {"html": "https://wpnews.pro/news/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world", "markdown": "https://wpnews.pro/news/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world.md", "text": "https://wpnews.pro/news/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world.txt", "jsonld": "https://wpnews.pro/news/arxiv-endorsement-request-for-cs-lg-diagnostic-control-for-hierarchical-world.jsonld"}}