{"author":{"@type":"Person","name":"Alexander Bock","url":"https://abxk.ai/en/autor/alexander-bock/"},"categories":["Physical AI"],"content":{"markdown":"In Level 1 of this learning path, we tested whether a computer can tell from an EEG which hand a person imagines moving. The answer was sobering: the decoder was right {{\u003c num \"eeg\" \"eeg_acc_median_pct\" \u003e}}% of the time for the median person, where guessing gives 50%, and only {{\u003c num \"eeg\" \"eeg_over70_n\" \u003e}} of {{\u003c num \"eeg\" \"eeg_subjects\" \u003e}} people reached the 70% that BCI research often names as the minimum for useful control. So what would make such a decoder better?\n\nThis is part 5 of Level 2 in our learning path Physical AI. We ran four more experiments on the same public recordings: we changed the number of electrodes, the amount of training data and the source of the training data, and we compared imagined with real movements. You will learn how a decoder is built from simple steps, why more electrodes do not automatically help and why a decoder trained on strangers beat one trained on the person's own data.\n\n## How the decoder works\n\nA decoder for brain signals is a chain of simple steps. Each step throws away what does not help and keeps what does:\n\n| Step | What it does | In our decoder |\n|---|---|---|\n| Filter | Keeps the frequencies that matter | Mu and beta rhythms, 8 to 30 waves per second |\n| Window | Cuts out the moment of the task | 0.5 to 2.5 seconds after the cue |\n| Spatial filters | Combine electrodes into a few strong signals | Common spatial patterns, four signals |\n| Features | Turn each signal into a number | The strength of each signal |\n| Classifier | Decides left or right | Linear discriminant analysis |\n| Test | Measures accuracy on unseen attempts | Five-fold cross-validation |\n\n## More electrodes are not automatically better\n\nWe gave the decoder 2, 3, 21 or all 64 electrodes. With just the two electrodes over the hand areas, C3 and C4, the median person reached {{\u003c num \"eeg\" \"dec_c3c4_pct\" \u003e}}%. Adding the electrode in the middle raised it to {{\u003c num \"eeg\" \"dec_c3czc4_pct\" \u003e}}%, and the 21 electrodes over the motor cortex reached {{\u003c num \"eeg\" \"dec_motor21_pct\" \u003e}}%. All 64 electrodes, though, did not do better: {{\u003c num \"eeg\" \"dec_all64_pct\" \u003e}}%.\n\n![Bar chart of the median accuracy for four electrode sets: 2 electrodes about 58 percent, 3 electrodes about 60 percent, 21 electrodes over the motor cortex about 60 percent and all 64 electrodes about 59 percent; a line marks guessing at 50 percent](/images/learn/en/inside-a-brain-decoder-electrodes.webp \"Median accuracy by number of electrodes. The electrodes over the motor cortex carry the useful signal.\")\n\nThe reason is the small amount of data. Every extra electrode adds numbers that the decoder has to learn from the same few dozen attempts, and the more numbers it learns from little data, the more it adapts to noise instead of the real pattern. The electrodes over the motor cortex carry the useful signal; the others mostly add noise.\n\n## Training data matters\n\nWe also gave the decoder fewer of the person's own attempts to learn from. With only 10 attempts, the median accuracy was {{\u003c num \"eeg\" \"dec_train10_pct\" \u003e}}%, barely better than guessing. With 20 attempts, it was {{\u003c num \"eeg\" \"dec_train20_pct\" \u003e}}%, with 30 attempts {{\u003c num \"eeg\" \"dec_train30_pct\" \u003e}}%. Every extra minute of calibration helps, and a few dozen attempts are clearly not enough for most people.\n\n## Learning from other people's brains\n\nThat raises an obvious question: why not learn from other people? For each person, we trained the decoder on all attempts of the other 105 people, several thousand in total, and tested it on the person it had never seen. The result surprised us: the median accuracy was {{\u003c num \"eeg\" \"dec_cross_pct\" \u003e}}%, higher than with the person's own data. For {{\u003c num \"eeg\" \"dec_cross_better_n\" \u003e}} of {{\u003c num \"eeg\" \"eeg_subjects\" \u003e}} people, the decoder trained on strangers did better than the one trained on their own attempts, and {{\u003c num \"eeg\" \"dec_cross_over70_n\" \u003e}} people reached 70% or more.\n\n![Scatter chart with one dot per person: on the horizontal axis the accuracy of a decoder trained on the person's own attempts, on the vertical axis the accuracy of a decoder trained on 105 other people; about half of the dots lie above the diagonal, where the decoder from other people does better](/images/learn/en/inside-a-brain-decoder-transfer.webp \"Own data or other people's data? Each dot is one person; above the diagonal, the decoder from other people wins.\")\n\nMany attempts from many people give a stable picture of the typical pattern, which a few dozen own attempts cannot. But brains differ: for about half of the people, their own data was still better. Research therefore combines both, starting with a model trained on many people and adapting it to the new user with a few minutes of their own data.\n\n## Real movement instead of imagination\n\nThe recordings also contain runs in which the same people really opened and closed their fists. With these, the median accuracy was {{\u003c num \"eeg\" \"dec_move_pct\" \u003e}}%, only slightly higher than for imagined movements, and {{\u003c num \"eeg\" \"dec_move_better_n\" \u003e}} of {{\u003c num \"eeg\" \"dec_move_n\" \u003e}} people were decoded better when they moved. For this decoder, imagining a movement and making it turned out to be surprisingly similar. That is good news for people with paralysis, for whom imagination is all they have.\n\n{{\u003c data \"eeg-decoder\" \u003e}}\n\n## What this means for brain-computer interfaces\n\n- **Calibration is the bottleneck.** Every new user needs training data, and our results show that a few dozen attempts are too few. Models trained on many people can shorten the setup.\n- **The right place beats many electrodes.** A small cap over the motor cortex can work as well as a full one, and it is faster to put on.\n- **Practice helps both sides.** The decoder learns from the user, and with feedback, the user learns to produce clearer patterns.\n- **The physics stays.** No amount of data removes the blurring of the skull, which is why implants remain far ahead for complex tasks.\n\n## How we ran the experiment\n\nWe used the EEG Motor Movement/Imagery Dataset from the BCI2000 project on PhysioNet, the same {{\u003c num \"eeg\" \"eeg_subjects\" \u003e}} people as in Level 1, with three runs of imagined and three runs of real fist movements per person. All variants use the same chain of steps; the electrode sets differ only in the channels they use. For the amount of training data, we drew 10 random splits per person and size; for other people's data, we trained on all other people and tested on the person left out. All random choices use a fixed seed.","text":"In Level 1 of this learning path, we tested whether a computer can tell from an EEG which hand a person imagines moving. The answer was sobering: the decoder was right 58.9% of the time for the median person, where guessing gives 50%, and only 31 of 106 people reached the 70% that BCI research often names as the minimum for useful control. So what would make such a decoder better?\nThis is part 5 of Level 2 in our learning path Physical AI. We ran four more experiments on the same public recordings: we changed the number of electrodes, the amount of training data and the source of the training data, and we compared imagined with real movements. You will learn how a decoder is built from simple steps, why more electrodes do not automatically help and why a decoder trained on strangers beat one trained on the person\u0026rsquo;s own data.\nHow the decoder works A decoder for brain signals is a chain of simple steps. Each step throws away what does not help and keeps what does:\nInside a Brain Decoder: Electrodes, Training Data and Other People Step What it does In our decoder Filter Keeps the frequencies that matter Mu and beta rhythms, 8 to 30 waves per second Window Cuts out the moment of the task 0.5 to 2.5 seconds after the cue Spatial filters Combine electrodes into a few strong signals Common spatial patterns, four signals Features Turn each signal into a number The strength of each signal Classifier Decides left or right Linear discriminant analysis Test Measures accuracy on unseen attempts Five-fold cross-validation More electrodes are not automatically better We gave the decoder 2, 3, 21 or all 64 electrodes. With just the two electrodes over the hand areas, C3 and C4, the median person reached 57.8%. Adding the electrode in the middle raised it to 59.9%, and the 21 electrodes over the motor cortex reached 60%. All 64 electrodes, though, did not do better: 58.9%.\nMedian accuracy by number of electrodes. The electrodes over the motor cortex carry the useful signal. The reason is the small amount of data. Every extra electrode adds numbers that the decoder has to learn from the same few dozen attempts, and the more numbers it learns from little data, the more it adapts to noise instead of the real pattern. The electrodes over the motor cortex carry the useful signal; the others mostly add noise.\nTraining data matters We also gave the decoder fewer of the person\u0026rsquo;s own attempts to learn from. With only 10 attempts, the median accuracy was 52.4%, barely better than guessing. With 20 attempts, it was 53.6%, with 30 attempts 55.3%. Every extra minute of calibration helps, and a few dozen attempts are clearly not enough for most people.\nLearning from other people\u0026rsquo;s brains That raises an obvious question: why not learn from other people? For each person, we trained the decoder on all attempts of the other 105 people, several thousand in total, and tested it on the person it had never seen. The result surprised us: the median accuracy was 62.2%, higher than with the person\u0026rsquo;s own data. For 54 of 106 people, the decoder trained on strangers did better than the one trained on their own attempts, and 32 people reached 70% or more.\nOwn data or other people\u0026#39;s data? Each dot is one person; above the diagonal, the decoder from other people wins. Many attempts from many people give a stable picture of the typical pattern, which a few dozen own attempts cannot. But brains differ: for about half of the people, their own data was still better. Research therefore combines both, starting with a model trained on many people and adapting it to the new user with a few minutes of their own data.\nReal movement instead of imagination The recordings also contain runs in which the same people really opened and closed their fists. With these, the median accuracy was 60%, only slightly higher than for imagined movements, and 58 of 106 people were decoded better when they moved. For this decoder, imagining a movement and making it turned out to be surprisingly similar. That is good news for people with paralysis, for whom imagination is all they have.\nWhat makes the decoder better or worse (all participants) VariantRecognized correctly (median)People with 70% or more 2 electrodes (C3, C4)57.8%16 3 electrodes (C3, Cz, C4)59.9%28 21 electrodes over the motor cortex60.0%33 All 64 electrodes58.9%31 Only 10 own attempts to learn from52.4%10 20 own attempts to learn from53.6%16 30 own attempts to learn from55.3%23 Learned from 105 other people62.2%32 Real movement instead of imagination60.0%28 Source as for the first table on this page.\nWhat this means for brain-computer interfaces Calibration is the bottleneck. Every new user needs training data, and our results show that a few dozen attempts are too few. Models trained on many people can shorten the setup. The right place beats many electrodes. A small cap over the motor cortex can work as well as a full one, and it is faster to put on. Practice helps both sides. The decoder learns from the user, and with feedback, the user learns to produce clearer patterns. The physics stays. No amount of data removes the blurring of the skull, which is why implants remain far ahead for complex tasks. How we ran the experiment We used the EEG Motor Movement/Imagery Dataset from the BCI2000 project on PhysioNet, the same 106 people as in Level 1, with three runs of imagined and three runs of real fist movements per person. All variants use the same chain of steps; the electrode sets differ only in the channels they use. For the amount of training data, we drew 10 random splits per person and size; for other people\u0026rsquo;s data, we trained on all other people and tested on the person left out. All random choices use a fixed seed.\n","words":989},"date":"2026-10-18","description":"What makes a decoder for brain signals better: more electrodes, more training, other people's data or real movement? Four tests on EEG recordings of 106 people.","image":"https://abxk.ai/images/cover/inside-a-brain-decoder.webp","json":"https://abxk.ai/en/physical-ai/learn/inside-a-brain-decoder/index.json","kind":"page","language":"en","learning_path":{"category":"physical-ai","level":2,"level_name":"Intermediate"},"modified":"2026-10-18","section":"physical-ai","seo_title":"What Makes a Brain Decoder Work: An EEG Experiment","tags":["eeg","brain-computer-interface","decoder","transfer-learning","experiment"],"title":"Inside a Brain Decoder: Electrodes, Training Data and Other People","translations":[{"json":"https://abxk.ai/de/physical-ai/learn/inside-a-brain-decoder/index.json","language":"de","url":"https://abxk.ai/de/physical-ai/learn/inside-a-brain-decoder/"}],"url":"https://abxk.ai/en/physical-ai/learn/inside-a-brain-decoder/"}