AI Neuroscience Concept

Blind Spots Uncovered on the Intersection of AI and Neuroscience – Dozens of Scientific Papers Debunked

AI Neuroscience Concept

Findings debunk dozens of outstanding revealed papers claiming to learn minds with EEG.

Is it attainable to learn an individual’s thoughts by analyzing the electrical indicators from the mind? The reply could also be way more complicated than most individuals suppose.

Purdue College researchers – working on the intersection of synthetic intelligence and neuroscience – say a outstanding dataset used to attempt to reply this query is confounded, and subsequently many eye-popping findings that have been based mostly on this dataset and obtained high-profile recognition are false in spite of everything.

The Purdue staff carried out in depth checks over multiple yr on the dataset, which appeared on the mind exercise of people collaborating in a research the place they checked out a sequence of photographs. Every particular person wore a cap with dozens of electrodes whereas they seen the photographs.

The Purdue staff’s work is revealed in IEEE Transactions on Sample Evaluation and Machine Intelligence. The staff obtained funding from the Nationwide Science Basis.

EEG Cap With Electrodes

Purdue College researchers are doing work on the intersection of synthetic intelligence and neuroscience. On this photograph, a analysis participant is sporting an EEG cap with electrodes. Credit score: Chris Adam/Purdue College

“This measurement approach, often called electroencephalography or EEG, can present details about mind exercise that would, in precept, be used to learn minds,” mentioned Jeffrey Mark Siskind, professor {of electrical} and pc engineering in Purdue’s Faculty of Engineering. “The issue is that they used EEG in a manner that the dataset itself was contaminated. The research was performed with out randomizing the order of photographs, so the researchers have been in a position to inform what picture was being seen simply by studying the timing and order info contained in EEG, as an alternative of fixing the actual downside of decoding visible notion from the mind waves.”

The Purdue researchers initially started questioning the dataset after they couldn’t acquire comparable outcomes from their very own checks. That’s after they began analyzing the earlier outcomes and decided {that a} lack of randomization contaminated the dataset.

“This is likely one of the challenges of working in cross-disciplinary analysis areas,” mentioned Hari Bharadwaj, an assistant professor with a joint appointment in Purdue’s Faculty of Engineering and Faculty of Well being and Human Sciences. “Essential scientific questions usually demand cross-disciplinary work. The catch is that, typically, researchers educated in a single area are usually not conscious of the widespread pitfalls that may happen when making use of their concepts to a different. On this case, the prior work appears to have suffered from a disconnect between AI/machine-learning scientists, and pitfalls which are well-known to neuroscientists.”

The Purdue staff reviewed publications that used the dataset for duties comparable to object classification, switch studying and technology of photographs depicting human notion and thought utilizing brain-derived representations measured by electroencephalograms (EEGs)

“The query of whether or not somebody can learn one other individual’s thoughts by electrical mind exercise may be very legitimate,” mentioned Ronnie Wilbur, a professor with a joint appointment in Purdue’s Faculty of Well being and Human Sciences and Faculty of Liberal Arts. “Our analysis exhibits that a greater strategy is required.”

Reference: “The Perils and Pitfalls of Block Design for EEG Classification Experiments” by Ren Li, Jared S. Johansen, Hamad Ahmed, Thomas V. Ilyevsky, Ronnie B. Wilbur, Hari M. Bharadwaj and Jeffrey Mark Siskind, 19 November 2020, IEEE Transactions on Sample Evaluation and Machine Intelligence.
DOI: 10.1109/TPAMI.2020.2973153

Siskind is a well known Purdue innovator and has labored on a number of patented applied sciences with the Purdue Analysis Basis Workplace of Expertise Commercialization.

Source link

Leave a Comment

Your email address will not be published. Required fields are marked *