File size: 3,425 Bytes
efb9dfc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
{
  "schema_version": 1,
  "title": "Reproduction: Maximum Likelihood Reinforcement Learning",
  "emoji": "🎯",
  "space_id": "Umong/repro-maximum-likelihood-reinforcement-learning",
  "paper": {
    "arxiv_id": "2602.02710"
  },
  "tags": [
    "icml2026-repro",
    "paper-EeuLO2BjFN"
  ],
  "updated_at": "2026-07-18T18:54:07+00:00",
  "root": {
    "slug": "index",
    "title": "Reproduction: Maximum Likelihood Reinforcement Learning",
    "file": "pages/index.md",
    "children": [
      {
        "slug": "executive-summary",
        "title": "Executive summary",
        "file": "pages/executive-summary/page.md",
        "children": []
      },
      {
        "slug": "claim-1-maxrl-defines-a-compute-indexed-family-of-sample-based-objectives-that-interpolates-between-standard-rl-and-exact-maximum-likelihood-as-sampling-compute-increases-abstract",
        "title": "Claim 1: MaxRL defines a compute-indexed family of sample-based objectives that interpolates between standard RL and exact maximum likelihood as sampling compute increases (Abstract).",
        "file": "pages/claim-1-maxrl-defines-a-compute-indexed-family-of-sample-based-objectives-that-interpolates-between-standard-rl-and-exact-maximum-likelihood-as-sampling-compute-increases-abstract/page.md",
        "children": []
      },
      {
        "slug": "claim-2-the-maxrl-objectives-admit-a-simple-unbiased-policy-gradient-estimator-for-non-differentiable-sampling-settings-abstract",
        "title": "Claim 2: The MaxRL objectives admit a simple unbiased policy-gradient estimator for non-differentiable sampling settings (Abstract).",
        "file": "pages/claim-2-the-maxrl-objectives-admit-a-simple-unbiased-policy-gradient-estimator-for-non-differentiable-sampling-settings-abstract/page.md",
        "children": []
      },
      {
        "slug": "claim-3-the-paper-claims-maxrl-converges-to-maximum-likelihood-optimization-in-the-infinite-compute-limit-abstract",
        "title": "Claim 3: The paper claims MaxRL converges to maximum-likelihood optimization in the infinite-compute limit (Abstract).",
        "file": "pages/claim-3-the-paper-claims-maxrl-converges-to-maximum-likelihood-optimization-in-the-infinite-compute-limit-abstract/page.md",
        "children": []
      },
      {
        "slug": "claim-4-empirically-maxrl-pareto-dominates-tested-existing-methods-across-all-evaluated-models-and-tasks-abstract",
        "title": "Claim 4: Empirically, MaxRL Pareto-dominates tested existing methods across all evaluated models and tasks (Abstract).",
        "file": "pages/claim-4-empirically-maxrl-pareto-dominates-tested-existing-methods-across-all-evaluated-models-and-tasks-abstract/page.md",
        "children": []
      },
      {
        "slug": "claim-5-maxrl-reports-up-to-20x-test-time-scaling-efficiency-gains-compared-with-a-grpo-trained-counterpart-abstract",
        "title": "Claim 5: MaxRL reports up to 20x test-time scaling efficiency gains compared with a GRPO-trained counterpart (Abstract).",
        "file": "pages/claim-5-maxrl-reports-up-to-20x-test-time-scaling-efficiency-gains-compared-with-a-grpo-trained-counterpart-abstract/page.md",
        "children": []
      },
      {
        "slug": "conclusion",
        "title": "Conclusion",
        "file": "pages/conclusion/page.md",
        "children": []
      }
    ]
  },
  "agent_view_tokens": 3376,
  "revision": "1784400847404442706"
}