Profile cover photo
Profile photo
Francisco Pereira
brain boffin, machine learning mercenary
brain boffin, machine learning mercenary
Francisco's posts

Post has attachment
An interesting story that illustrates all too well how cognitive biases and incomplete information can lead to rather adverse outcomes. Also, a 1979 Soviet attempt to use a computer to inform decision making and put the biases in the open.

"A nuclear weapons command exercise by NATO in November 1983 prompted fear in the leadership of the Soviet Union that the maneuvers were a cover for a nuclear surprise attack by the United States, triggering a series of unparalleled Soviet military re­sponses. (...)
The U.S. intelligence community, the board said, “did not at the time, and for several years afterwards, attach sufficient weight to the possibility that the war scare was real.” (...)
The PFIAB review repeatedly criticized U.S. intelligence on Soviet leaders, saying at the time of the 1984 post­­mortems that “the US knew very little about Kremlin decision­making.” It added, “Our own leadership needs far better intelligence reporting on and assessments of the mind­set of the Soviet leadership — its ideological/political instincts and perceptions.” And the review said the 1984 estimates “were over­confident, particularly in the judgments pertaining to Soviet leadership intentions — since little intelligence, human or technical, existed to support them.” (...)
“During my first years in Washington,” Reagan said, “I think many of us in the administration took it for granted that the Russians, like ourselves, considered it unthinkable that the United States would launch a first strike against them. But the more experience I had with the Soviet leaders and other heads of state who knew them, the more I began to realize that many Soviet officials feared us not only as adversaries but as potential aggressors who might hurl nuclear weapons at them in a first strike; because of this, and perhaps because of a sense of insecurity and paranoia with roots reaching back to the invasions of Russia by Napoleon and Hitler, they had aimed a huge arsenal of nuclear weapons at us."

Post has attachment
A gripping article telling an intellectual sleuthing story connecting David Hume to Buddhism, with a poignant personal angle, from a researcher whose papers and books you may well have read...

Post has attachment
"Abandoning more than a century of clinical research, theoretical developments, and observational studies, psychologists worldwide announced that their entire professional lives had been utterly worthless, as the human brain could never comprehend its own workings, let alone understand its own understanding."

Post has attachment
+Ke Yang, here's something to look up to!

A Research Specialist position is available in Matthew Botvinick's lab in the Princeton Neuroscience Institute, in collaboration with Francisco Pereira at Siemens Corporate Technology. This is a position within a team participating in the IARPA Knowledge Representation in Neural Systems program. The goal of the program is to develop models of how the brain represents conceptual knowledge, and how that knowledge is used when carrying out tasks like reading a sentence.  These models will be tested using brain imaging data. 

The main role in this position is as a programmer writing the code needed to build such models, while also being involved in experiment design and data analysis. Examples of tasks you might be asked to do: 

- prepare text corpora for use in developing models 
- implement evaluation tasks that the models developed will be benchmarked with 
- collect material from online resources and behavioral experiments 
- prepare stimuli for brain imaging experiments or model building 
- help in designing and carrying out brain imaging and behavioral experiments 
- analyze behavioral and brain imaging experimental data 

It is an unusual position in that you will be gaining experience in both machine learning and cognitive neuroscience. In addition to the core research goals, we will also be delivering a concrete system to the funding agency, and hence this will be a fast-paced project. You will be working directly with the PIs and also graduate students and research scientists in the team.  
Essential qualifications are experience developing software in MATLAB or Python (Perl is also useful), as well as an undergraduate degree in computer science, biomedical engineering, cognitive neuroscience or a related field. Preferred qualifications are knowledge in natural language processing, machine learning, and collection, preprocessing and analysis of brain imaging data (e.g. SPM, FSL or AFNI). The final candidate will be required to pass a background check successfully.  
This is a 1 year position, starting as soon as possible, with an additional year possible contingent on funding availability and performance. To apply, please go to
and use “Search Open Positions” with requisition number 1400452.
Please email with any pre-application inquiries.
Wait while more posts are being loaded