DreamCode · Multimodal decoding of dream content from EEG signals through AI

Dreams offer a unique window into conscious experience during sleep, yet scientific access to their content remains extremely limited. Traditional methods rely on verbal reports after awakening, which are prone to forgetting, distortion and bias. Recent advances in neuroscience and artificial intelligence now enable a transformative shift in how dreams can be studied. This project, DreamCode, aims to develop and validate a novel, non-invasive method to decode the content of dreams directly from brain activity. DremCode uses EEG recordings during sleep, combined with individualized decoding models and state-of-the-art multimodal computational models that capture the structure of memory in a fine-grained, high-dimensional format. By linking neural data with rich semantic representations spanning language, vision and sound, the project will aim at reconstructing dream content. Testing will involve healthy individuals exposed to controlled stimuli prior to sleep, followed by EEG-based analysis and comparison with dream reports. By establishing reliable, personalized mappings between brain signals and structured semantic representations, DreamCode aims to deliver the first objective and scalable pipeline for dream decoding. If validated, this approach will redefine empirical dream research and open new pathways for clinical applications: from trauma and nightmare therapy to the diagnosis of psychiatric and neurological conditions involving altered dream states, such as depression and sleep behavior disorder.