Neuronal codes for visual perception and memory
Introduction
One of the most important and at the same time fascinating functions of the brain is to construct meaning from relatively little, noisy and ambiguous sensory information. We indeed see, and are aware of seeing, a dog, a laptop, a famous actor or our mother, rather than the minute features that characterize these images or the combination of wavelengths that impinge our retina. The notion that our sensations and thoughts are based on constructions we make of the external world goes all the way back to the beginnings of scientific reasoning in the thoughts of Aristotle (Aristotle, reprinted 2004) and has been particularly substantiated by Hermann von Helmholz, who argued that perception is given by representations made from unconscious inferences (Helmholtz, 1878). However, perception is just the beginning of elaborate brain processes that are triggered by seeing, for example, the face of an old friend in a cafe. In particular, recognizing the face of our friend will elicit the recall of related information and it will eventually lead to the formation of memories about the new encounter.
In the following, I will focus on visual perception-given that this is the dominant sensory modality in humans and many other animals, and also given the vast number of studies dealing with vision compared to other sensory modalities – although I would argue that the principles described below apply not only to vision but to sensory processing in general. A myriad of studies have characterized the neural substrates of visual perception along the so-called ventral visual pathway (VVP) (Grill-Spector and Malach, 2004, Logothetis and Sheinberg, 1996, Roelfsema, 2006, Tanaka, 1996; Tsao and Livingstone, 2008). From high level visual areas in the VVP, there are strong connections to the medial temporal lobe (MTL) (see Fig. 1) (Lavenex and Amaral, 2000, Saleem and Tanaka, 1996, Suzuki, 1996), where evidence from patient H.M. (Corkin, 2002, Milner et al., 1968, Scoville and Milner, 1957; Squire, 2009), other patients with lesions in the MTL (Moscovitch et al., 2005, Rosenbaum et al., 2005; Squire et al., 2004) and animal studies (Mishkin, 1978; Squire and Zola-Morgan, 1991; Squire et al., 2004) have demonstrated the key role of this area in declarative memory (Squire and Zola-Morgan, 1991; Squire et al., 2004). In spite of the strong relationship between perception and memory, both behaviorally (considering that perception triggers memory formation) and anatomically (given that high level VVP areas project to the MTL), these two processes have been mostly studied independently (for previous studies linking perception and memory see Carlson et al., 2013, Clarke et al., 2013, Cowell et al., 2010, Miyashita, 1993, Naya et al., 2001, Quian Quiroga, 2012b), without focusing on the processes and the subjacent neuronal representations that determine how external stimuli elicits the creation of the conceptual representations that we use, for example, to form new episodic memories. (In the following, by “concept” I refer to the meaning of a stimulus – e.g. a given person, place or object-as opposed to a representation of sensory features.) These are precisely the two processes – visual perception in the VVP and memory in the MTL – that I will discuss and contrast in this short review. In particular, I will argue that both perception and memory involve a construction of meaning by selecting little information and using inferences and abstractions. But while perception involves distributed representations, with category responses and a hierarchical organization of information that is ideal for the robust identification of a stimulus and its related semantic information, (episodic) memory involves much sparser, invariant and not organized representations that are better suited for the fast creation of associations.
The review is structured as follows: I will first outline basic similarities between perception and memory processes, then I will briefly describe perception along the VVP – just highlighting a few selected studies, given that this topic has already been comprehensively reviewed elsewhere (see references above) – and in the final two sections I will describe more in detail the findings with single cell recordings in the MTL and, based on this type of representation, I will then discuss a very simple model of memory formation based on partial overlapping assemblies.
Section snippets
Perception and memory
Although perception and memory are two different processes, they are intrinsically related because the perception of a person or an object requires contrasting sensory features with stored representations. In fact, cases of visual associative agnosia provide clear evidence of this link considering that, due to specific lesions, these patients cannot associate a percept with its meaning (i.e. they cannot access the memory of the percept), in spite of having normal vision (Farah and Feinberg, 2000
Perception in the ventral visual pathway
Given the vast evidence about visual perception along the ventral visual pathway reviewed elsewhere (Grill-Spector and Malach, 2004, Logothetis and Sheinberg, 1996, Roelfsema, 2006, Tanaka, 1996; Tsao and Livingstone, 2008), in this section I will just summarize evidence from electrophysiology recordings in monkeys and humans about the encoding of stimuli in visual cortex, which I will later contrast with evidence from recordings in the human MTL. In particular, I will argue that in visual
Single cell recordings in the human medial temporal lobe
The hippocampus and its surrounding cortex, what is known as the medial temporal lobe, is involved in certain forms of epilepsy and it is therefore a typical area targeted for the implantation of intracranial electrodes (Engel et al., 2005, Niedermeyer, 1993, Rey et al., 2015). The MTL comprises several interconnected areas organized in a hierarchical structure (Fig. 1B): the perirhinal and parahippocampal cortex (PHC), at the bottom of this hierarchy, receive inputs from cortical sensory areas
Partially overlapping assemblies
In line with the large bulk of evidence demonstrating the role of the MTL in declarative memory (Moscovitch et al., 2005; Squire and Zola-Morgan, 1991; Squire et al., 2004), we have postulated that concept cells are critically involved in the formation and recall of episodic memories by encoding associations between concepts (Quian Quiroga, 2012b). The representation by concept cells seems indeed optimal for this function, as we tend to remember concepts and forget myriads of irrelevant
Contrasting cortical and MTL neuronal representations
In the previous sections we have argued that along the areas involved in processing sensory stimuli there is an increase of invariance and selectivity to complex features. High level visual areas project to the MTL, where this process continues further and reaches its pinnacle in the hippocampus, with neurons firing selectively to different pictures and even the names of specific individuals or objects. Although the general process of increasing invariance and selectivity to complex
Conclusions and open questions
In this review I have described and contrasted basic principles of neural coding in cortex and the MTL, arguing that while distributed and organized representations in cortex are ideal for perception and storage of semantic information, the non-organized and sparse representations in the MTL are ideal for the formation and recall of episodic memories. Moreover, I have described a model of how associations can be rapidly created in the MTL based on partially overlapping assemblies. Such
Acknowledgements
I thank Hernan Rey, Joaquin Navajas and Theofanis Panagiotaropoulos for useful comments on earlier versions of the manuscript and the Human Frontiers Science Program (grant nr. RGP0015/2013) for funding. I also thank Bruno Rossion (the action editor) and an anonymous reviewer for very useful and detailed feedback that helped improving this contribution.
References (193)
- et al.
Analysis of extracellular firing patterns of deep temporal lobe structures in man
Electroencephalogr. Clin. Neurophysiol.
(1973) - et al.
The time course of visual processing: backward masking and natural scene categorisation
Vis. Res.
(2005) - et al.
Beyond faces and modularity: the power of an expertise framework
Trends Cognit. Sci.
(2006) - et al.
Human hippocampal neurons predict how well word Paris will be remembered
Neuron
(2001) - et al.
Experimental and theoretical approaches to conscious processing
Neuron
(2011) - et al.
The hippocampus, memory, and place cells: is it spatial memory or a memory space?
Neuron
(1999) - et al.
Neural “ignition”: enhanced activation linked to perceptual awareness in human ventral stream visual cortex
Neuron
(2009) - et al.
Single neuron activity in human hippocampus and amygdala during recognition of faces and objects
Neuron
(1997) - et al.
Implicit learning of geometric eigenfaces
Vis. Res.
(2014) - et al.
Separate visual pathways for perception and action
Trends Neurosci.
(1992)