Search

This article offers a formal account of curiosity and insight in terms of active (Bayesian) inference. It deals with the dual problem of inferring states of the world and learning its statistical structure. In contrast to current trends in machine learning (e.g., deep learning), we focus on how people attain insight and understanding using just a handful of observations, which are solicited through curious behavior. We use simulations of abstract rule learning and approximate Bayesian inference to show that minimizing (expected) variational free energy leads to active sampling of novel contingencies. This epistemic behavior closes explanatory gaps in generative models of the world, thereby reducing uncertainty and satisfying curiosity. We then move from epistemic learning to model selection or structure learning to show how abductive processes emerge when agents test plausible hypotheses about symmetries (i.e., invariances or rules) in their generative models. The ensuing Bayesian model reduction evinces mechanisms associated with sleep and has all the hallmarks of "aha" moments. This formulation moves toward a computational account of consciousness in the pre-Cartesian sense of sharable knowledge (i.e., con: "together"; scire: "to know"). [ABSTRACT FROM AUTHOR]

The activity that takes place in memory systems during sleep is likely to be related to the role of sleep in memory consolidation and learning, as well as to the generation of dream hallucinations. This study addressed the often-stated hypothesis that replay of whole episodic memories contributes to the multimodal hallucinations of sleep. Over a period of 14 days, 29 subjects kept a log of daytime activities, events, and concerns, wrote down any recalled dreams, and scored the dreams for incorporation of any waking experiences. While 65% of a total of 299 sleep mentation reports were judged to reflect aspects of recent waking life experiences, the episodic replay of waking events was found in no more than 1 – 2% of the dream reports. This finding has implications for understanding the unique memory processing that takes place during the night and is consistent with evidence that sleep has no role in episodic memory consolidation. [ABSTRACT FROM AUTHOR]

Performance on a visual discrimination task shows long-term improvement after a single training session. When tested within 24 hr of training, improvement was not observed unless subjects obtained at least 6 hr of post-training sleep prior to retesting, in which case improvement was proportional to the amount of sleep in excess of 6 hr. For subjects averaging 8 hr of sleep, overnight improvement was proportional to the amount of slow wave sleep (SWS) in the first quarter of the night, as well as the amount of rapid eye movement sleep (REM) in the last quarter. REM during the intervening 4 hr did not appear to contribute to improvement. A two-step process, modeling throughput as the product of the amount of early SWS and late REM, accounts for 80 percent of intersubject variance. These results suggest that, in the case of this visual discrimination task, both SWS and REM are required to consolidate experience-dependent neuronal changes into a form that supports improved task performance. [ABSTRACT FROM AUTHOR]

The notion that dreaming might alter the strength of associative links in memory was first proposed almost 200 years ago. But no strong evidence of such altered associative links has been obtained. Semantic priming can be used to quantify the strength of associative links between pairs of words; it is thought to measure the automatic spread of activation from a "node" representing one word to nodes representing semantically related words. Semantic priming could thus be used to test for global alterations in the strengths of associative links across the wake-sleep cycle. Awakenings from REM and nonREM (NREM) sleep produce a period of state carry-over during which performance is altered as a result of the brain's slow transition to full wakefulness, and cognitive testing in this period can provide information about the functioning of the brain during the prior sleep period. When subjects were tested across the night - before and after a night's sleep as well as immediately following forced awakenings from REM and NREM sleep-weak priming (e.g., thief-wrong) was found to be state dependent (p = 0.016), whereas strong priming (e.g., hot-cold) was not (p = 0.89). Weak primes were most effective in the presleep and REM sleep conditions and least effective in NREM and postsleep conditions. Most striking are analyses comparing weak and strong priming within each wake-sleep state. Contrary to the normal pattern of priming, subjects awakened from REM sleep showed greater priming by weak primes than by strong primes (p = 0.01). This result was seen in each of three protocols. In contrast, strong priming exceeded weak priming in NREM sleep. The shift in weak priming seen after REM sleep awakenings suggests that cognition during REM sleep is qualitatively different from that of waking and NREM sleep and may reflect a shift in associative memory systems, a shift that we hypothesize underlies the bizarre and hyperassociative character of REMsleep dreaming. Known changes in brainstem activity that control the transition into and maintenance of REM sleep provide a possible explanation of this shift. [ABSTRACT FROM AUTHOR]