A high-capacity model for one shot association learning in the brain

Inverse forgetting in unconscious episodic memory

Forming memories of experienced episodes calls upon the episodic memory system. Episodic encoding may proceed with and without awareness of episodes. While up to 60% of consciously encoded episodes are forgotten after 10 h, the fate of unconsciously encoded episodes is unknown. Here we track over 10 h, which are filled with sleep or daytime activities, the retention of unconsciously and consciously experienced episodes. The episodes were displayed in cartoon clips that were presented weakly and strongly masked for conscious and unconscious encoding, respectively. Clip retention was tested for distinct clips directly after encoding, 3 min and 10 h after encoding using a forced-choice test that demands deliberate responses in both consciousness conditions. When encoding was conscious, retrieval accuracy decreased by 25% from 3 min to 10 h, irrespective of sleep or wakefulness. When encoding was unconscious, retrieval accuracy increased from 3 min to 10 h and depended on sleep. Hence, opposite to the classic forgetting curve, unconsciously acquired episodic memories strengthen over time and hinge on sleep on the day of learning to gain influence over human behavior.

Larger capacity for unconscious versus conscious episodic memory

Episodic memory is the memory for experienced events. A peak competence of episodic memory is the mental combination of events to infer commonalities. Inferring commonalities may proceed with and without consciousness of events. Yet what distinguishes conscious from unconscious inference? This question inspired nine experiments that featured strongly and weakly masked cartoon clips presented for unconscious and conscious inference. Each clip featured a scene with a visually impenetrable hiding place. Five animals crossed the scene one-by-one consecutively. One animal trajectory represented one event. The animals moved through the hiding place, where they might linger or not. The participants' task was to observe the animals' entrances and exits to maintain a mental record of which animals hid simultaneously. We manipulated information load to explore capacity limits. Memory of inferences was tested immediately, 3.5 or 6 min following encoding. The participants retrieved inferences well when encoding was conscious. When encoding was unconscious, the participants needed to respond intuitively. Only habitually intuitive decision makers exhibited a significant delayed retrieval of inferences drawn unconsciously. Their unconscious retrieval performance did not drop significantly with increasing information load, while conscious retrieval performance dropped significantly. A working memory network, including hippocampus, was activated during both conscious and unconscious inference and correlated with retrieval success. An episodic retrieval network, including hippocampus, was activated during both conscious and unconscious retrieval of inferences and correlated with retrieval success. Only conscious encoding/retrieval recruited additional brain regions outside these networks. Hence, levels of consciousness influenced the memories' behavioral impact, memory capacity, and the neural representational code.

Multiassociative Memory: Recurrent Synapses Increase Storage Capacity

The connection density of nearby neurons in the cortex has been observed to be around 0.1, whereas the longer-range connections are present with much sparser density (Kalisman, Silberberg, & Markram, 2005 ). We propose a memory association model that qualitatively explains these empirical observations. The model we consider is a multiassociative, sparse, Willshaw-like model consisting of binary threshold neurons and binary synapses. It uses recurrent synapses for iterative retrieval of stored memories. We quantify the usefulness of recurrent synapses by simulating the model for small network sizes and by doing a precise mathematical analysis for large network sizes. Given the network parameters, we can determine the precise values of recurrent and afferent synapse densities that optimize the storage capacity of the network. If the network size is like that of a cortical column, then the predicted optimal recurrent density lies in a range that is compatible with biological measurements. Furthermore, we show that our model is able to surpass the standard Willshaw model in the multiassociative case if the information capacity is normalized per strong synapse or per bits required to store the model, as considered in Knoblauch, Palm, and Sommer ( 2010 ).