[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-mojo-unlabeled-training-neural-decoding-en":3,"article-related-mojo-unlabeled-training-neural-decoding-en":30,"series-research-10ffae7d-4474-45a7-87ee-d7a3f348c5de":73},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"10ffae7d-4474-45a7-87ee-d7a3f348c5de","mojo-unlabeled-training-neural-decoding-en","MOJO adds unlabeled training to neural decoding","\u003Cp data-speakable=\"summary\">MOJO mixes masked autoencoding with supervised training to make spike-tokenizing neural decoders work better with little labeled data.\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>Research org\u003C\u002Fstrong>: Unspecified in arXiv abstract\u003C\u002Fli>\u003Cli>\u003Cstrong>Core data\u003C\u002Fstrong>: No benchmark numbers in abstract\u003C\u002Fli>\u003Cli>\u003Cstrong>Breakthrough\u003C\u002Fstrong>: Joint masked autoencoding and supervised training for spike-tokenizing models\u003C\u002Fli>\u003C\u002Ful>\u003Cp>Neural decoders are only useful if they stay accurate when data gets messy, sparse, or comes from a new session. That matters for brain-computer interfaces, closed-loop experiments, and any workflow that has to turn neural activity into predictions without collecting huge labeled datasets every time.\u003C\u002Fp>\u003Cp>This paper argues that the usual supervised-only recipe leaves performance on the table. The authors introduce \u003Ca href=\"https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.14086\">MOJO (Masked autOencoder-based JOint training)\u003C\u002Fa>, a training framework that lets spike-tokenizing models learn from both labeled and unlabeled neural data.\u003C\u002Fp>\u003Ch2>What problem this paper is trying to fix\u003C\u002Fh2>\u003Cp>The core issue is simple: many neural decoding systems need paired behavior labels to train, but those labels are expensive and often limited. In real deployments, you may have plenty of neural recordings and very little task annotation, especially when moving to a new session, a new subject, or a new recording setup.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784181773896-pot3.png\" alt=\"MOJO adds unlabeled training to neural decoding\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>The abstract says recent spike-level tokenization has already helped with multi-session pretraining and decoding quality. But those models are still restricted to supervised learning, which means they cannot directly use unlabeled data. That restriction is a problem if you want a decoder that generalizes instead of overfitting to one narrow labeled dataset.\u003C\u002Fp>\u003Cp>MOJO is designed to close that gap. Instead of treating self-supervised learning and supervised learning as separate tracks, it combines them in one training framework for spike-tokenizing models.\u003C\u002Fp>\u003Ch2>How MOJO works in plain English\u003C\u002Fh2>\u003Cp>The method blends two objectives. One is supervised learning, which still teaches the model to predict the target behavior. The other is self-supervised learning through masked autoencoding, which trains the model to reconstruct missing parts of the neural input.\u003C\u002Fp>\u003Cp>That masked reconstruction step matters because it gives the model a reason to learn structure in the neural data even when no label is attached. In practice, that means the model can make use of recordings that would otherwise be ignored during supervised training.\u003C\u002Fp>\u003Cp>Because the framework is built around spike-tokenizing models, it fits into the same kind of representation learning pipeline that has already been useful for multi-session neural decoding. The difference is that MOJO does not stop at labeled examples. It uses unlabeled data as a first-class training signal.\u003C\u002Fp>\u003Cp>There is also an interpretability angle here. The abstract says the added SSL objective improves neuronal representations enough to help with brain region classification and spike-statistics prediction, even though those tasks were not explicitly optimized for.\u003C\u002Fp>\u003Ch2>What the paper actually shows\u003C\u002Fh2>\u003Cp>The authors evaluate MOJO on three spiking datasets: monkey motor cortex during reaching, and multi-regional mouse recordings during vision and decision-making tasks. They also test it on human electrocorticography during speech, which is important because that pushes the method beyond spikes into a different neural modality.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784181771064-eznc.png\" alt=\"MOJO adds unlabeled training to neural decoding\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Across those settings, the abstract says MOJO outperforms purely supervised-trained models. The strongest gains show up when labeled data is scarce, especially in few-shot finetuning where only a small amount of labeled data from a new session is available.\u003C\u002Fp>\u003Cp>The paper also claims that MOJO generalizes beyond spiking data and reaches performance comparable to neuro-\u003Ca href=\"\u002Fnews\u002Fdatabricks-query-foundation-models-guide-en\">foundation models\u003C\u002Fa> designed specifically for continuous signals. That is a meaningful result because it suggests the training idea is not locked to one recording format.\u003C\u002Fp>\u003Cp>What the abstract does not give is a table of exact \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> numbers. So if you are looking for direct percentage gains, error reductions, or task-specific scores, those details are not available in the provided summary. The takeaway is directional but clear: adding SSL improves performance, especially when labels are limited.\u003C\u002Fp>\u003Cul>\u003Cli>Three datasets were used for spiking data evaluation.\u003C\u002Fli>\u003Cli>The method was also tested on human electrocorticography during speech.\u003C\u002Fli>\u003Cli>The biggest gains were reported in few-shot finetuning and label-impoverished settings.\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>Why developers and neural engineers should care\u003C\u002Fh2>\u003Cp>If you build decoding systems, this paper points to a practical training strategy: do not throw away unlabeled neural data. In domains like BCI and closed-loop neuroscience, unlabeled recordings are often much easier to collect than clean behavior annotations, so a method that can exploit them is immediately useful.\u003C\u002Fp>\u003Cp>It also suggests a path toward more flexible foundation-style training for neural data. The abstract explicitly frames the result as a step toward more scalable data usage when training neuro-foundation models, which is a useful signal for teams trying to unify pretraining and downstream decoding.\u003C\u002Fp>\u003Cp>For engineers, the most interesting part is not just better accuracy. It is the training recipe itself: masked autoencoding plus supervised objectives, applied to spike-tokenizing models, and extended to another modality without collapsing. That combination makes the approach feel more like a reusable systems pattern than a one-off model tweak.\u003C\u002Fp>\u003Ch2>Limitations and open questions\u003C\u002Fh2>\u003Cp>The abstract is positive, but it still leaves important questions open. It does not provide benchmark numbers in the summary, so the scale of the gains is unclear from the source alone.\u003C\u002Fp>\u003Cp>It also does not spell out training cost, \u003Ca href=\"\u002Ftag\u002Finference\">inference\u003C\u002Fa> cost, or whether the joint objective adds meaningful complexity compared with supervised-only training. For real deployments, those details matter as much as raw accuracy.\u003C\u002Fp>\u003Cp>Another open question is how broadly the approach transfers beyond the datasets named in the abstract. The paper shows three spiking datasets plus one human ECoG setting, which is promising, but that is still a limited slice of the full neural decoding landscape.\u003C\u002Fp>\u003Cp>Even with those caveats, the contribution is straightforward: unlabeled neural data is not just extra data, it is a useful signal. MOJO formalizes that idea in a way that fits current spike-tokenizing pipelines and appears to improve generalization when labels are scarce.\u003C\u002Fp>\u003Cp>For teams working on neural interfaces, that is the main lesson. If your decoder needs to survive new sessions, new subjects, or new recording conditions, a training loop that mixes supervised and self-supervised objectives may be a better default than supervised learning alone.\u003C\u002Fp>","MOJO mixes masked autoencoding with supervised training to make spike-tokenizing neural decoders work better with little labeled data.","arxiv.org","https:\u002F\u002Farxiv.org\u002Fabs\u002F2607.14086",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784181773896-pot3.png","research","en","faf69af0-1ce7-49a4-a9c2-4a0a3afb6065",[17,18,19,20,21],"neural decoding","self-supervised learning","masked autoencoding","brain-computer interfaces","spike tokenization",[23,24,25],"MOJO combines masked autoencoding with supervised training for spike-tokenizing neural models.","The method improves decoding most when labeled data is limited, especially in few-shot finetuning.","It also transfers to human ECoG and improves representation quality on auxiliary tasks.",1,"2026-07-16T06:02:25.088864+00:00","2026-07-16T06:02:25.067+00:00","96fcb5cf-c4ed-4e7b-9615-8b8f55347631",{"tags":31,"relatedLang":32,"relatedPosts":36},[],{"id":15,"slug":33,"title":34,"language":35},"mojo-unlabeled-training-neural-decoding-zh","MOJO 用無標註資料強化神經解碼","zh",[37,43,49,55,61,67],{"id":38,"slug":39,"title":40,"cover_image":41,"image_url":41,"created_at":42,"category":13},"20f1341a-8779-4c17-a8f0-1274b3ad2de6","metaperch-metadata-bioacoustics-foundation-models-en","MetaPerch uses metadata to boost bioacoustics models","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784185413772-docp.png","2026-07-16T07:02:33.987049+00:00",{"id":44,"slug":45,"title":46,"cover_image":47,"image_url":47,"created_at":48,"category":13},"196a6f70-fce8-4e6f-b6a9-be1f4459541b","ot-ica-wasserstein-linear-ica-en","OT-ICA Uses Wasserstein Distance for Linear ICA","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784183594908-2dd1.png","2026-07-16T06:32:39.335324+00:00",{"id":50,"slug":51,"title":52,"cover_image":53,"image_url":53,"created_at":54,"category":13},"a169cb71-ab29-462b-b773-ba2a7dce52aa","claude-j-space-is-not-consciousness-en","Claude's J-space is not consciousness, but it matters","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784100798014-3un1.png","2026-07-15T07:32:39.232247+00:00",{"id":56,"slug":57,"title":58,"cover_image":59,"image_url":59,"created_at":60,"category":13},"220d77ea-3f5a-4e96-b557-997a01727119","terrazero-zero-demo-self-play-driving-en","TerraZero trains driving agents with no demos","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784099000554-m5xc.png","2026-07-15T07:02:41.693527+00:00",{"id":62,"slug":63,"title":64,"cover_image":65,"image_url":65,"created_at":66,"category":13},"d040e868-c97f-4620-b267-89a20683dfd4","seriality-gap-video-diffusion-models-en","The Seriality Gap in Video Diffusion Models","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784097179835-yh7l.png","2026-07-15T06:32:36.985476+00:00",{"id":68,"slug":69,"title":70,"cover_image":71,"image_url":71,"created_at":72,"category":13},"d781f88c-c0b0-4467-a8cf-2b47b126f946","e3-ai-agents-task-complexity-en","E3 Helps AI Agents Stop Over-Reading Simple Tasks","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784095395632-exrl.png","2026-07-15T06:02:36.486816+00:00",[74,79,84,89,94,99,104,109,114,119],{"id":75,"slug":76,"title":77,"created_at":78},"a2715e72-1fe8-41b3-abb1-d0cf1f710189","ai-predictions-2026-big-changes-en","AI Predictions for 2026: Brace for Big Changes","2026-03-26T01:25:07.788356+00:00",{"id":80,"slug":81,"title":82,"created_at":83},"8404bd7b-4c2f-4109-9ec4-baf29d88af2b","ml-papers-of-the-week-github-research-desk-en","ML Papers of the Week Turns GitHub Into a Research Desk","2026-03-27T01:11:39.480259+00:00",{"id":85,"slug":86,"title":87,"created_at":88},"87897a94-8065-4464-a016-1f23e89e17cc","ai-ml-conferences-to-watch-in-2026-en","AI\u002FML Conferences to Watch in 2026","2026-03-27T01:51:54.184108+00:00",{"id":90,"slug":91,"title":92,"created_at":93},"6f1987cf-25f3-47a4-b3e6-db0997695be8","openclaw-agents-manipulated-self-sabotage-en","OpenClaw Agents Can Be Manipulated Into Failure","2026-03-28T03:03:18.899465+00:00",{"id":95,"slug":96,"title":97,"created_at":98},"a53571ad-735a-4178-9f93-cb09b699d99c","vega-driving-language-instructions-en","Vega: Driving with Natural Language Instructions","2026-03-28T14:54:04.698882+00:00",{"id":100,"slug":101,"title":102,"created_at":103},"a34581d6-f36e-46da-88bb-582fb3e7425c","personalizing-autonomous-driving-styles-en","Drive My Way: Personalizing Autonomous Driving Styles","2026-03-28T14:54:26.148181+00:00",{"id":105,"slug":106,"title":107,"created_at":108},"2bc1ad7f-26ce-4f02-9885-803b35fd229d","training-knowledge-bases-writeback-rag-en","Training Knowledge Bases with WriteBack-RAG","2026-03-28T14:54:45.643433+00:00",{"id":110,"slug":111,"title":112,"created_at":113},"71adc507-3c54-4605-bbe2-c966acd6187e","packforcing-long-video-generation-en","PackForcing: Efficient Long-Video Generation Method","2026-03-28T14:55:02.646943+00:00",{"id":115,"slug":116,"title":117,"created_at":118},"675942ef-b9ec-4c5f-a997-381250b6eacb","pixelsmile-facial-expression-editing-en","PixelSmile Framework Enhances Facial Expression Editing","2026-03-28T14:55:20.633463+00:00",{"id":120,"slug":121,"title":122,"created_at":123},"6954fa2b-8b66-4839-884b-e46f89fa1bc3","adaptive-block-scaled-data-types-en","IF4: Smarter 4-Bit Quantization That Adapts to Your Data","2026-03-31T06:00:36.65963+00:00"]