Loading
Life sciences · Preprint
arXiv · September 10, 2026
Early or partial results. Treat as a signal, not a conclusion.
This preprint reports a controlled computational study showing that Mixture-of-Experts models degrade more rapidly than dense transformers when training data is repeated, with sparsity amplifying overfitting. Regularization methods such as dropout and masking can partially mitigate this effect, and the authors identify early routing stabilization and expert over-specialization as correlated mechanisms. The work has not been peer reviewed and remains exploratory; the findings are limited to controlled training dynamics in a data repetition setting and do not directly validate performance on held-out or downstream tasks.
Controlled ablation study: computational experiments varying data repetition rate, model architecture (dense vs. MoE), expert count, and regularization method. Language models (Transformers, Mixture-of-Experts); no human subjects or real-world deployment.. Intervention: Data repetition at rates from 1× to 64×; regularization methods including dropout and masking-based regularization applied to MoE and dense Transformer models. Compared with: Dense Transformer models of equivalent active parameter count; MoE models without regularization.
80M dense models tolerate 8× data repetition with minimal degradation, whereas MoEs begin degrading at 4× repetition MoEs underperform dense models after 32× data repetition, losing their all-unique data performance advantage With strong masking-based regularization, MoEs outperform dense models even at 64× data repetition
Safety was not reported in the material analysed. Check the source before drawing any conclusion about harm.
The source did not state who this applies to in practice.
Single-site computational study on language model training dynamics with controlled experimental conditions but no peer review, limited to controlled synthetic settings, and mechanistic rather than applied validation.
As stated by the source record.
Quoted from the source exactly as published.
Graded across the dimensions that decide whether you should act, each from what the source actually supports. There is no single score, and where a dimension was not assessed it says so.
As the supply of human-written text is exhausted, it has become standard practice to repeat language model training data. Prior work has studied data repetition for densely activated Transformers, but the effects of data repetition remains largely unexplored for recently dominant sparse architectures such as Mixture-of-Experts (MoE), despite their increased compute efficiency. We vary data repetition rates across single- and multi-domain data mixes, and across MoE settings, including expert count and granularity. We consistently find, for models ranging from 80M to 1B active (8.5B total) parameters, that MoEs degrade more rapidly under data repetition. This effect increases with sparsity, dictated by total rather than active parameters. While 80M dense models can repeat data over 8x with minimal degradation, MoEs instead begin to suffer at 4x, and deteriorate rapidly, ceding their performance benefits in all-unique data settings to underperform dense models after 32x. We experiment with existing regularization methods as a potential remedy. We find that some methods, such as dropout, can mitigate overfitting. In particular, with strong masking-based regularization, MoEs are able to outperform dense models even when data is repeated more than 64 times. However, no method fully matches the performance of all-unique training data. Finally, we analyze internal mechanisms correlated with MoE overfitting in high repetition regimes, and find that MoE routing universally stabilizes early in training, and that expert specialization correlates with overfitting to repeated data. In sum, our work addresses the adverse interactions between sparsity and data repetition: we present evidence for the core mechanisms of overfitting and its potential remediation, and suggest promising avenues for future methods to reduce over-specialization in model parameters by disrupting memorization patterns.
Taken from the source record, never inferred. Follow any of these and new work involving them reaches your briefing.