{"id":"21015ac6-25f7-44c8-9eda-7d0df5bc1f17","arxiv_id":"2606.25975","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Tensorion generalizes Muon to tensors via a tractable LMO on a tensor norm ball, recovers Muon exactly for matrices, and reports improved convergence over Adam and tensor baselines in evaluated CV settings.","lead":"Tensorion extends the Muon optimizer from matrices to higher-order tensors using a linear minimization oracle over a chosen tensor norm ball that reduces to unfolding matrices. A smart generalist might read it for potential gains in training stability for models with multilinear parameter structures common in computer vision.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest_assumption correctly isolates the key technical requirement. Because the provided text supplies no equations, proofs, or counter-examples that would show the bound fails or the LMO is mis-derived, the assumption stands as the load-bearing point but is not shown to be violated. Hence no adjustment to the UNVERDICTED status is warranted on the basis of an identifiable flaw.","tokens_in":1690,"tokens_out":271,"duration_ms":15638,"concrete_test":"Implement the claimed LMO for a small random 3-tensor, compute the resulting step, and compare its norm against the true tensor spectral norm (via power iteration on the unfoldings); if the ratio exceeds 1 + ε for small ε the bound is not tight.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central construction rests on selecting a tensor norm whose ball yields an LMO reducible to adaptive unfoldings while still bounding the multilinear spectral norm. The abstract states this balance is achieved and that the matrix case recovers Muon exactly. No internal inconsistency, missing derivation step, or experimental setup flaw is visible from the given material that would falsify the claim that such a norm exists and is useful.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces Tensorion as a generalization of the Muon optimizer from matrices to higher-order tensors. It constructs a tensor norm ball whose linear minimization oracle (LMO) is claimed to be tractable by reduction to operations on adaptively selected unfolding matrices while still tightly bounding the tensor spectral norm; the matrix case recovers Muon exactly, and experiments on tensor-based computer vision tasks are said to show improved convergence and more stable updates versus Adam and existing tensor-aware baselines.","tokens_in":1747,"tokens_out":282,"duration_ms":18320,"significance":"If the norm construction, LMO reduction, and empirical gains hold, the work would extend principled spectral-norm constrained optimization to the multilinear setting common in modern models, addressing a clear limitation of vector-based methods like Adam.","major_comments":[{"comment":"Abstract (and throughout): no derivation or explicit definition of the tensor norm is supplied, nor is there a proof that the chosen norm both bounds the multilinear spectral norm and yields a tractable LMO via adaptive unfoldings; without these steps the central technical claim cannot be verified.","section":"Abstract"},{"comment":"Abstract: the experimental claim of improved convergence and stability is stated without any quantitative results, protocol details, dataset descriptions, or baseline implementations, rendering the performance assertion impossible to assess.","section":"Abstract"}],"minor_comments":[],"recommendation":"uncertain","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their comments on our manuscript. We address each major comment point by point below.","responses":[{"response":"The abstract serves as a high-level summary and therefore omits the full technical details. The manuscript body supplies the explicit definition of the tensor norm, the derivation showing that it tightly bounds the multilinear spectral norm, and the proof of the tractable LMO reduction to adaptive unfolding matrices. We will revise the abstract to include a concise reference to these elements and their guarantees.","revision_made":"yes","referee_comment":"[Abstract] Abstract (and throughout): no derivation or explicit definition of the tensor norm is supplied, nor is there a proof that the chosen norm both bounds the multilinear spectral norm and yields a tractable LMO via adaptive unfoldings; without these steps the central technical claim cannot be verified."},{"response":"Space limitations in the abstract preclude inclusion of quantitative results or full protocol details. The manuscript's Experiments section provides the complete experimental protocol, dataset descriptions for the evaluated computer vision tasks, baseline implementations, and quantitative results showing the reported improvements. We will revise the abstract to add a brief statement of key quantitative findings.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the experimental claim of improved convergence and stability is stated without any quantitative results, protocol details, dataset descriptions, or baseline implementations, rendering the performance assertion impossible to assess."}],"tokens_in":1274,"tokens_out":322,"duration_ms":30424,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core move here is lifting Muon's spectral-norm steepest descent to higher-order tensors by defining a norm ball whose linear minimization oracle reduces to operations on adaptively chosen unfolding matrices. When the input is a matrix the construction collapses back to Muon, which is a clean consistency check.\n\nThe paper does a reasonable job framing the practical requirement: the norm has to control the tensor spectral norm without making the LMO intractable. If the reduction works as described, it could be a usable drop-in for models whose weights are already multilinear.\n\nThe experiments are only summarized as showing better convergence and more stable updates than Adam and existing tensor baselines on some computer-vision tensor tasks. That direction is the right one to test, but the abstract gives no numbers, baselines, or protocol, so the strength of the result is still open.\n\nThe soft spot is the norm itself. The text says it is chosen to balance tight spectral control with tractability, yet supplies neither the explicit definition nor the argument that the bound remains useful. Without that step the central claim rests on an unshown construction. The same goes for any complexity or stability analysis of the unfolding procedure.\n\nThis is for people already following Muon-style constrained optimizers and looking for tensor extensions. A reader in that niche would want to see the derivation and the actual runs.\n\nIt is worth sending to a serious referee so the math and the empirical claims can be checked directly.","headline":"Tensorion extends Muon to tensors with an LMO that reduces to adaptive unfoldings and recovers the matrix case exactly, but the norm construction and experimental evidence need the full details to assess.","tokens_in":2239,"tokens_out":378,"would_cite":false,"duration_ms":19456,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Tensorion generalizes Muon from matrices to higher-order tensors via a linear minimization oracle over a tensor norm ball.","keywords":["Tensorion","Muon optimizer","tensor-aware optimization","linear minimization oracle","spectral norm","unfolding matrices","machine learning optimizers","tensor parameters"],"falsifier":"An experiment in which Tensorion updates on tensor vision tasks show no gain in convergence speed or stability over Adam baselines, or in which the oracle steps violate the intended spectral norm bound.","tokens_in":2596,"feed_emoji":"","tokens_out":596,"duration_ms":40432,"temperature":0.7,"pith_summary":"The paper introduces Tensorion to handle the multilinear structure of tensor parameters in machine learning models instead of treating them as flat vectors. It extends Muon's spectral-norm steepest descent to tensors by defining a norm whose linear minimization oracle reduces to operations on selected unfolding matrices. The design keeps the oracle tractable while aiming to bound the tensor spectral norm. A reader would care because many models use tensor weights, and respecting that structure could change how gradients are updated during training. When limited to matrices the method matches Muon exactly.","feed_headline":"Tensorion generalizes Muon optimizer to tensors","feed_subtitle":"A linear minimization oracle on unfolding matrices handles higher-order weights and yields more stable updates than Adam in tested vision ta","key_machinery":"The linear minimization oracle over a tensor norm ball chosen to bound the spectral norm while reducing to unfolding-matrix operations.","core_discovery":"Tensorion performs constrained optimization over a tensor norm ball through its linear minimization oracle. The norm is selected so the oracle remains computable by reducing to operations on adaptively chosen unfolding matrices while still bounding the tensor spectral norm. The method recovers Muon exactly on order-2 tensors. On tensor-based computer vision tasks it produces more stable gradient updates and improved convergence relative to Adam and existing tensor-aware methods.","pith_inferences":["The same unfolding reduction might extend to other structured parameter spaces such as those appearing in physics or recommendation models.","Different norm choices could trade bound tightness against computation speed in future variants.","Testing the method on non-vision tensor problems would show whether the gains depend on the specific task structure."],"forward_implications":["Tensorion recovers Muon exactly when restricted to matrices.","It applies to higher-order tensor weights in computer vision models.","It yields more stable gradient updates than Adam-based methods in the tested settings.","The oracle stays tractable for tensors beyond order 2 through unfolding reductions."],"fun_headline_variants":["Tensorion generalizes Muon via tensor norm ball LMO","Tensorion reduces to Muon for order-2 tensors","LMO on unfolding matrices enables Tensorion","Tensor-aware optimizer bounds spectral norm via unfoldings"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"A tensor norm exists that bounds the spectral norm tightly enough for the linear minimization oracle to stay tractable through operations on unfolding matrices.","fun_headline_variants_meta":{"raw":{"variants":["Tensorion generalizes Muon via tensor norm ball LMO","Tensorion reduces to Muon for order-2 tensors","LMO on unfolding matrices enables Tensorion","Tensor-aware optimizer bounds spectral norm via unfoldings"]},"model":"grok-4.3","cost_usd":0.00632,"raw_usage":{"total_tokens":2957,"prompt_tokens":643,"num_sources_used":0,"completion_tokens":61,"cost_in_usd_ticks":63199500,"prompt_tokens_details":{"text_tokens":643,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2253,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":643,"tokens_out":61,"duration_ms":18617,"temperature":1.0,"reasoning_tokens":2253,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T19:59:52.535953+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"An experiment in which Tensorion updates on tensor vision tasks show no gain in convergence speed or stability over Adam baselines, or in which the oracle steps violate the intended spectral norm bound.","supporting_citations":[],"review_version":1}