Created at 6am, Mar 27
Ms-RAGScience
0
Desiderata of evidence for representation in neuroscience
i6G5ruG6smP2fj9-wQdtD9dYNCUWMS3Qqt5pvxbyObc
File Type
PDF
Entry Count
173
Embed. Model
jina_embeddings_v2_base_en
Index Type
hnsw

Stephan POHL§, Department of Philosophy, New York University, stephan.pohl@nyu.eduEdgar Y WALKER, Department of Physiology and Biophysics, Computational NeuroscienceCenter, University of Washington, Seattle, WA, eywalker@uw.eduDavid L BARACK, Departments of Neuroscience and Philosophy, University of Pennsylvania,Philadelphia, PA, dbarack@gmail.comJennifer LEE, Center for Neural Science, New York University, jll616@nyu.eduRachel N DENISON, Department of Psychological & Brain Sciences, Boston University, Boston,MA, rdenison@bu.eduNed BLOCK, Department of Philosophy, New York University, ned.block@nyu.eduFlorent MEYNIEL#, Cognitive Neuroimaging Unit, INSERM, CEA, CNRS, UniversitéParis-Saclay, NeuroSpin center, Gif/Yvette, France & Institute for Neuromodulation, GHU Parispsychiatrie et neuroscience, Sainte Anne Hospital, Paris, France, florent.meyniel@cea.frWei Ji MA# Center for Neural Science and Department of Psychology, New York University,weijima@nyu.edu#: co-senior authors; §: corresponding authorWhen claiming that a neural response represents a feature of the world, scientists try to establish that the neural response is (1) sensitive and (2) specific to the feature, (3) invariant to other features, and (4) functional downstream in the brain. We formalize these desiderata in information-theoretic terms, permitting their precise statement while unifying the different analysis methods used in neuroscience under one framework, thereby providing a common terminology to researchers. We discuss how common analysis methods are used to evaluate the desiderata and present canonical examples to illustrate the desiderata at work in research practice.This paper develops a systematic framework for the evidence neuroscientists use to establish whether a neural response represents a feature. Researchers try to establish that the neural response is (1) sensitive and (2) specific to the feature, (3) invariant to other features, and (4) functional, which means that it is used downstream in the brain. We formalize these desiderata in information-theoretic terms. This formalism allows us to precisely state the desiderata while unifying the different analysis methods used in neuroscience under one framework. We discuss how common methods such as correlational analyses, decoding and encoding models, representational similarity analysis, and tests of statistical dependence are used to evaluate the desiderata. In doing so, we provide a common terminology to researchers that helps to clarify disagreements, to compare and integrate results across studies and research groups, and to identify when evidence might be missing and when evidence for some representational conclusion is strong. We illustrate the framework with several canonical examples, including the representation of orientation, numerosity, faces, and spatial location. We end by discussing how the framework can be extended to cover models of the neural code, multi-stage models, and other domains.

4.1. Orientation of visual elements In their series of pioneering works, Hubel and Wiesel (1959, 1961, 1962, 1968) identified that the activities of cat and macaque V1 neurons are strongly modulated by the orientation of a moving edge as the edge slides in and out of the spatial receptive field of the neurons. By establishing that many V1 neurons have a preferential response to specific orientations, as commonly characterized by their orientation tuning curves, their work showed that V1 carries information about the stimulus orientation, thereby establishing sensitivity and specificity. Furthermore, their work established two distinct types of orientation-sensitive neurons in V1simple cells and complex cellsby showing that the responses of complex cells demonstrated invariance to the spatial phase of an oriented stimulus within the cells receptive field.
id: ae4f8c1a8f0040a61fe2b85888722d84 - page: 28
Orientation tuning curvesthe average response of a neuron to each stimulus orientationhave been one of the earliest and most influential encoding models in which the variability of V1 neuron responses is explained in terms of the variability in the stimulus orientation (Sompolinsky & Shapley, 1997; Victor, Purpura, Katz, & Mao, 1994); the performance of such models has been evaluated based on the total variance explained, an approximate measure of specificity. Sensitivity of V1 has been characterized further with decoding models (both linear and nonlinear) in order to establish the decodability of stimulus orientation from the V1 population activity (Berens et al., 2012; Chen, Geisler, & Seidemann, 2006, 2008). By manipulating additional variables, such work has also shown that V1 exhibits conditional sensitivity and conditional specificity to orientation, as conditioned on contrast (Anderson, Lampl, Gillespie, & Ferster, 2000; Berens et al., 2012; Hansel & van Vreeswijk, 2002; Nowak &
id: 80b7c66a73d4b136635d495bfa09b3d3 - page: 28
Barone, 2009), temporal frequency (Moore, Alitto, & Usrey, 2005), and spatial frequency (Jeon, Swain, Good, Chase, & Kuhlman, 2018). However, while V1 demonstrates a high degree of sensitivity, both conditioned and unconditioned, most V1 neurons exhibit low invariance to contrast, spatial frequency and temporal frequency. Hence in the face of variations in orientation along with other variables, V1 will typically exhibit low specificity to the stimulus orientation.
id: e5dd1e006580ac6b4cc63c3b6d9288e6 - page: 28
28 The functionality of V1 for orientation has been tested in the context of orientation discrimination tasks. When the animal is presented with ambiguous orientation stimuli, apparently random activity in V1 has been shown to be predictive of the animals decision above chancea phenomenon known as choice probability (Nienborg & Cumming, 2014). While this work does not clearly establish a causal connection, being able to predict behavior from the neural response at least establishes functional specificity. More recently, deep neural network models have been used to predict the animals orientation-based decision from V1 activity even when conditioned on the stimulus, again helping to establish functional conditional specificity (Walker, Cotton, Ma, & Tolias, 2020). 4.2. Numerosity
id: ae8dafc13d43c3a0aff81138bcbb76df - page: 28
How to Retrieve?
# Search

curl -X POST "https://search.dria.co/hnsw/search" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"rerank": true, "top_n": 10, "contract_id": "i6G5ruG6smP2fj9-wQdtD9dYNCUWMS3Qqt5pvxbyObc", "query": "What is alexanDRIA library?"}'
        
# Query

curl -X POST "https://search.dria.co/hnsw/query" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"vector": [0.123, 0.5236], "top_n": 10, "contract_id": "i6G5ruG6smP2fj9-wQdtD9dYNCUWMS3Qqt5pvxbyObc", "level": 2}'