{"id":"c718c89b-7615-4b4f-8c06-0ab19a7d2472","arxiv_id":"2606.10130","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":3.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Compares pruning methods (Variance Filter, Branch and Bound) and regularization (LASSO, ridge) for training physical reservoir computers, reporting gains on nonlinear benchmarks using a fiber-optical extreme learning machine.","lead":"This paper tests output pruning strategies and L1/L2 regularization to reduce overfitting and computation when training optical reservoir computers on nonlinear tasks. A smart generalist might read it to see practical ways to make physical computing systems more efficient without heavy retraining.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.3","headline":"Generalizability of pruning/regularization gains rests on untested assumption that fiber-optical ELM results transfer to other physical substrates.","rationale":"Reader's weakest_assumption directly identifies the same platform-transfer risk; full-text review confirms no additional cross-substrate validation or formal argument mitigates it, so the provisional UNVERDICTED stance remains appropriate.","tokens_in":1706,"tokens_out":312,"duration_ms":12142,"concrete_test":"Apply the same Equal Search, Branch-and-Bound, Variance Filter, LASSO, and ridge procedures to an independent physical reservoir (e.g., a memristor array or optoelectronic delay-line system) on the identical Spiral Benchmark; if relative gains over baseline fall below 50% of those reported for the fiber ELM, the general claim does not hold.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central empirical claims (readout selection across full spectrum improves non-iterative methods; L1/L2 regularization boosts nonlinear tasks like Spiral Benchmark) are demonstrated only on a single nonlinear fiber-optical extreme learning machine. The manuscript states the methods are 'of general use' but provides no cross-platform experiments, no theoretical derivation showing platform-independence, and no ablation on how substrate-specific dynamics (e.g., fiber nonlinearity vs. other photonic or electronic reservoirs) affect the observed pruning/regularization benefits. This leaves the transferability assumption as the least secure link in the argument that these training principles apply to physical reservoirs broadly.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript examines output pruning strategies (Equal Search, Branch and Bound, Variance Filter, and random pruning) and L1/L2 regularization (LASSO and ridge regression) to reduce overfitting and computational cost when training physical reservoir computers. Using a nonlinear fiber-optical extreme learning machine as the experimental platform, it reports that informed readout selection across the full output spectrum benefits non-iterative methods and that regularization improves performance on nonlinear benchmarks such as the Spiral task; the authors state that the methods are of general use while noting the results are obtained exemplarily on this substrate.","tokens_in":1821,"tokens_out":368,"duration_ms":13273,"significance":"If the reported performance gains are reproducible and the pruning/regularization benefits prove transferable, the work would supply concrete, practical guidelines for training physical reservoirs that could reduce overfitting and training overhead in optical and other hardware implementations.","major_comments":[{"comment":"Abstract: the central claim that the described methods 'significantly enhance performance' and are 'of general use' is presented without any quantitative metrics, error bars, dataset sizes, or baseline comparisons, rendering the empirical contribution impossible to evaluate from the provided text.","section":"Abstract"},{"comment":"Abstract and closing paragraph: the assertion that the pruning and regularization principles are 'of general use' for physical reservoirs rests on results from a single fiber-optical ELM substrate; no cross-platform experiments, no ablation of substrate-specific dynamics (e.g., fiber nonlinearity versus other photonic or electronic reservoirs), and no theoretical derivation establishing platform independence are supplied, leaving transferability as an untested assumption that directly supports the broad applicability claim.","section":"Abstract"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive comments on our manuscript. We address each major comment below and will revise the abstract accordingly to improve clarity and precision.","responses":[{"response":"We agree that the abstract would benefit from quantitative support. In the revised manuscript we will add specific metrics (e.g., accuracy improvements on the Spiral benchmark with LASSO/ridge regularization versus unregularized baselines, including standard deviations across repeated trials) together with dataset sizes and brief baseline comparisons. This will allow readers to evaluate the empirical contribution directly from the abstract.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the central claim that the described methods 'significantly enhance performance' and are 'of general use' is presented without any quantitative metrics, error bars, dataset sizes, or baseline comparisons, rendering the empirical contribution impossible to evaluate from the provided text."},{"response":"The manuscript already qualifies the claim by stating that 'results are obtained from and discussed exemplarily for a nonlinear fiber-optical extreme learning machine.' We acknowledge that no cross-platform experiments or theoretical derivation of platform independence are provided, as the work focuses on training principles rather than exhaustive validation across substrates. In revision we will further tone down the phrasing in the abstract and closing paragraph to 'principles expected to be of general use for physical reservoir computers, demonstrated here on a fiber-optical ELM' to avoid implying untested universality while retaining the practical motivation.","revision_made":"yes","referee_comment":"[Abstract] Abstract and closing paragraph: the assertion that the pruning and regularization principles are 'of general use' for physical reservoirs rests on results from a single fiber-optical ELM substrate; no cross-platform experiments, no ablation of substrate-specific dynamics (e.g., fiber nonlinearity versus other photonic or electronic reservoirs), and no theoretical derivation establishing platform independence are supplied, leaving transferability as an untested assumption that directly supports the broad applicability claim."}],"tokens_in":1347,"tokens_out":421,"duration_ms":17122,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core of this work is an empirical test of output pruning (variance filter versus search methods) and L1/L2 regularization on a nonlinear fiber-optical extreme learning machine. The authors show that informed readout selection across the spectrum helps non-iterative training and that regularization improves results on the Spiral Benchmark. They are explicit that the data come from one platform and treat the findings as examples.\n\nWhat stands out is the direct comparison of pruning strategies and the observation that variance filtering can be competitive with more expensive search while being simpler. The discussion of how shrinking the latent space affects sampling is also practical for hardware implementations.\n\nThe soft spot is the generality. The abstract and title frame the results as training principles for physical reservoirs, yet no other substrates are tested and no derivation shows why the benefits should be platform-independent. Fiber nonlinearity differs from, say, microring or electronic reservoirs, so the observed improvements could be tied to this specific dynamics. That leaves the transfer claim as an assumption rather than a demonstrated result.\n\nThe paper is useful for researchers already working with optical or photonic reservoirs who need concrete tuning advice. It is less relevant for readers looking for new theoretical principles or cross-hardware validation. The methods are standard and the experiments appear reproducible on their own terms.\n\nI would send it to peer review. The empirical comparisons are clear enough to be worth referee time, provided the authors adjust the scope language to match the single-platform evidence.","headline":"The paper applies standard ML pruning and regularization to one fiber-optical reservoir and reports gains on nonlinear tasks, but the broader claim of general training principles for physical reservoirs rests on an untested transfer assumption.","tokens_in":2321,"tokens_out":376,"would_cite":false,"duration_ms":13646,"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":"Selecting readouts across the full output spectrum plus L1 or L2 regularization reduces overfitting and raises accuracy in physical reservoir computers.","keywords":["reservoir computing","physical reservoirs","extreme learning machine","output pruning","regularization","optical computing","nonlinear dynamics","overfitting mitigation"],"falsifier":"A controlled test on a second physical reservoir (for example a photonic or spin-wave device) in which the same pruning rules and regularization strengths produce no accuracy gain or a clear drop on the Spiral Benchmark.","tokens_in":2617,"feed_emoji":"🔬","tokens_out":670,"duration_ms":12177,"temperature":0.7,"pith_summary":"The paper tests ways to train physical reservoir computers without overfitting by pruning the reservoir outputs and adding regularization. It compares search-based pruning methods against a variance-based filter and shows that keeping readouts distributed across the entire output spectrum helps most when training is non-iterative. L1 and L2 penalties are shown to improve results on strongly nonlinear benchmarks such as the Spiral task. All experiments use a fiber-optical extreme learning machine, yet the authors present the pruning and regularization steps as general tools for any physical reservoir.","feed_headline":"Full-spectrum readout selection plus L1/L2 cuts overfitting in physical reservoirs","feed_subtitle":"Pruning and regularization tested on a fiber-optical extreme learning machine improve accuracy on nonlinear benchmarks while lowering traini","key_machinery":"Output pruning (Equal Search, Branch and Bound, Variance Filter) combined with L1 (LASSO) and L2 (ridge) regularization applied to the readout weights of the physical reservoir.","core_discovery":"Enforcing readout selection across the full output spectrum improves performance, especially for non-iterative methods; L1 and L2 regularization significantly enhance performance on highly nonlinear tasks such as the Spiral Benchmark. These gains are obtained by comparing loss-minimizing search methods (Equal Search, Branch and Bound) against statistical filtering (Variance Filter) and random pruning, all applied to the output layer of a nonlinear fiber-optical extreme learning machine.","pith_inferences":["If the same pruning statistics hold across different physical media, a single set of selection rules could be ported between optical, electronic, and mechanical reservoirs.","The variance-filter approach may interact with the specific spectrum of the physical nonlinearity, suggesting a platform-dependent tuning step that the paper leaves open.","Extending the full-spectrum constraint to multi-layer or deep physical reservoirs could further reduce the need for iterative retraining."],"forward_implications":["Informed output sampling becomes essential once the latent space of the reservoir shrinks.","Regularization yields the largest lift precisely on tasks whose target function is highly nonlinear.","Non-iterative training methods gain more from full-spectrum readout selection than iterative ones.","Both pruning and regularization lower the computational cost of the training phase."],"fun_headline_variants":["Full-spectrum readout selection cuts overfitting in reservoirs","L1 L2 regularization enhances nonlinear task performance","Variance filter aids output selection in physical reservoirs","Full output spectrum sampling outperforms random pruning"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"That the fiber-optical extreme learning machine used for experiments is representative of physical reservoirs in general and that the observed gains from pruning and regularization will transfer to other physical substrates without platform-specific retuning.","fun_headline_variants_meta":{"raw":{"variants":["Full-spectrum readout selection cuts overfitting in reservoirs","L1 L2 regularization enhances nonlinear task performance","Variance filter aids output selection in physical reservoirs","Full output spectrum sampling outperforms random pruning"]},"model":"grok-4.3","cost_usd":0.007244,"raw_usage":{"total_tokens":3336,"prompt_tokens":662,"num_sources_used":0,"completion_tokens":53,"cost_in_usd_ticks":72437000,"prompt_tokens_details":{"text_tokens":662,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2621,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":662,"tokens_out":53,"duration_ms":18842,"temperature":1.0,"reasoning_tokens":2621,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-27T15:09:00.967147+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled test on a second physical reservoir (for example a photonic or spin-wave device) in which the same pruning rules and regularization strengths produce no accuracy gain or a clear drop on the Spiral Benchmark.","supporting_citations":[],"review_version":1}