scholarly journals Optimizing for generalization in the decoding of internally generated activity in the hippocampus

2016 ◽  
Author(s):  
Matthijs A.A. van der Meer ◽  
Alyssa A. Carey ◽  
Youki Tanaka
Keyword(s):  
Neural Activity ◽  
Cross Validation ◽  
Ground Truth ◽  
Place Cells ◽  
Memory Recall ◽  
Spike Density ◽  
Variable Space ◽  
Covert Activity ◽  
A Minor ◽  
Optimal Decoding

AbstractThe decoding of a sensory or motor variable from neural activity benefits from a known ground truth against which decoding performance can be compared. In contrast, the decoding of covert, cognitive neural activity, such as occurs in memory recall or planning, typically cannot be compared to a known ground truth. As a result, it is unclear how decoders of such internally generated activity should be configured in practice. We suggest that if the true code for covert activity is unknown, decoders should be optimized for generalization performance using cross-validation. Using ensemble recording data from hippocampal place cells, we show that this cross-validation approach results in different decoding error, different optimal decoding parameters, and different distributions of error across the decoded variable space. In addition, we show that a minor modification to the commonly used Bayesian decoding procedure, which enables the use of spike density functions, results in substantially lower decoding errors. These results have implications for the interpretation of covert neural activity, and suggest easy-to-implement changes to commonly used procedures across domains, with applications to hippocampal place cells in particular.

scholarly journals modelBuildR: an R package for model building and feature selection with erroneous classifications

PeerJ ◽  
2021 ◽  
Vol 9 ◽  
pp. e10849
Author(s):  
Maximilian Knoll ◽  
Jennifer Furkel ◽  
Juergen Debus ◽  
Amir Abdollahi
Keyword(s):  
Feature Selection ◽  
Cross Validation ◽  
Model Building ◽  
Linear Models ◽  
Binary Classification ◽  
Ground Truth ◽  
R Package ◽  
Methylation Array ◽  
Survival Difference ◽  
Error Probabilities

Background Model building is a crucial part of omics based biomedical research to transfer classifications and obtain insights into underlying mechanisms. Feature selection is often based on minimizing error between model predictions and given classification (maximizing accuracy). Human ratings/classifications, however, might be error prone, with discordance rates between experts of 5–15%. We therefore evaluate if a feature pre-filtering step might improve identification of features associated with true underlying groups. Methods Data was simulated for up to 100 samples and up to 10,000 features, 10% of which were associated with the ground truth comprising 2–10 normally distributed populations. Binary and semi-quantitative ratings with varying error probabilities were used as classification. For feature preselection standard cross-validation (V2) was compared to a novel heuristic (V1) applying univariate testing, multiplicity adjustment and cross-validation on switched dependent (classification) and independent (features) variables. Preselected features were used to train logistic regression/linear models (backward selection, AIC). Predictions were compared against the ground truth (ROC, multiclass-ROC). As use case, multiple feature selection/classification methods were benchmarked against the novel heuristic to identify prognostically different G-CIMP negative glioblastoma tumors from the TCGA-GBM 450 k methylation array data cohort, starting from a fuzzy umap based rough and erroneous separation. Results V1 yielded higher median AUC ranks for two true groups (ground truth), with smaller differences for true graduated differences (3–10 groups). Lower fractions of models were successfully fit with V1. Median AUCs for binary classification and two true groups were 0.91 (range: 0.54–1.00) for V1 (Benjamini-Hochberg) and 0.70 (0.28–1.00) for V2, 13% (n = 616) of V2 models showed AUCs < = 50% for 25 samples and 100 features. For larger numbers of features and samples, median AUCs were 0.75 (range 0.59–1.00) for V1 and 0.54 (range 0.32–0.75) for V2. In the TCGA-GBM data, modelBuildR allowed best prognostic separation of patients with highest median overall survival difference (7.51 months) followed a difference of 6.04 months for a random forest based method. Conclusions The proposed heuristic is beneficial for the retrieval of features associated with two true groups classified with errors. We provide the R package modelBuildR to simplify (comparative) evaluation/application of the proposed heuristic (http://github.com/mknoll/modelBuildR).

scholarly journals Learning from Weak-Label Data: A Deep Forest Expedition

2020 ◽  
Vol 34 (04) ◽  
pp. 6251-6258
Author(s):  
Qian-Wei Wang ◽  
Liang Yang ◽  
Yu-Feng Li
Keyword(s):  
Cross Validation ◽  
State Of The Art ◽  
Frequency Estimation ◽  
Ground Truth ◽  
Layer By Layer ◽  
Label Data ◽  
Information Layer ◽  
Deep Forest ◽  
Class Labels ◽  
Number Of Classes

Weak-label learning deals with the problem where each training example is associated with multiple ground-truth labels simultaneously but only partially provided. This circumstance is frequently encountered when the number of classes is very large or when there exists a large ambiguity between class labels, and significantly influences the performance of multi-label learning. In this paper, we propose LCForest, which is the first tree ensemble based deep learning method for weak-label learning. Rather than formulating the problem as a regularized framework, we employ the recently proposed cascade forest structure, which processes information layer-by-layer, and endow it with the ability of exploiting from weak-label data by a concise and highly efficient label complement structure. Specifically, in each layer, the label vector of each instance from testing-fold is modified with the predictions of random forests trained with the corresponding training-fold. Since the ground-truth label matrix is inaccessible, we can not estimate the performance via cross-validation directly. In order to control the growth of cascade forest, we adopt label frequency estimation and the complement flag mechanism. Experiments show that the proposed LCForest method compares favorably against the existing state-of-the-art multi-label and weak-label learning methods.

Responses of the teleostean nucleus isthmi to looming objects and other moving stimuli

2006 ◽  
Vol 23 (2) ◽  
pp. 209-219 ◽  
Author(s):  
SHAWN P. GALLAGHER ◽  
DAVID P.M. NORTHMORE
Keyword(s):  
Visual Field ◽  
Neural Activity ◽  
Three Dimensional ◽  
Object Motion ◽  
Activity Levels ◽  
Nucleus Isthmi ◽  
Spike Density ◽  
Ball Size ◽  
Longitudinal Body Axis ◽  
The Right

Visually evoked extracellular neural activity was recorded from the nucleus isthmi (NI) of goldfish and bluegill sunfish. When moving anywhere within the right eye's visual field, three-dimensional checkered balls or patterns on a computer screen evoked bursts of spikes in the left NI. Object motion parallel to the longitudinal body axis gave responses that habituated markedly upon repetition, but movement into recently unstimulated regions of the visual field gave vigorous responses. Thus, while NI's response is not visuotopic, its habituation is. An object approaching the animal's body generated a rising spike density, whereas object recession generated only a transient burst. During the approach of a checkered stimulus ball, average NI spike density rose linearly as the ball-to-eye distance decreased and at a rate proportional to the ball's speed (2.5–30 cm/s). Increasing ball size (2.2–9.2 cm) did not affect the rate of activity rise at a given speed, but did increase overall activity levels. NI also responded reliably to expanding textures of fixed overall size. The results suggest that NI signals changes in motion of objects relative to the fish, and estimates the proximity of approaching objects.

Composition and replay of mnemonic sequences: The contributions of REM and slow-wave sleep to episodic memory

2013 ◽  
Vol 36 (6) ◽  
pp. 610-611 ◽  
Author(s):  
Sen Cheng ◽  
Markus Werning
Keyword(s):  
Episodic Memory ◽  
Eye Movement ◽  
Slow Wave ◽  
Rem Sleep ◽  
Neural Activity ◽  
Slow Wave Sleep ◽  
Place Cells ◽  
Object Representations ◽  
Episodic Memories ◽  
Gamma Band Oscillations

AbstractWe propose that rapid eye movement (REM) and slow-wave sleep contribute differently to the formation of episodic memories. REM sleep is important for building up invariant object representations that eventually recur to gamma-band oscillations in the neocortex. In contrast, slow-wave sleep is more directly involved in the consolidation of episodic memories through replay of sequential neural activity in hippocampal place cells.

scholarly journals The Neural Correlates of Cognitive Reappraisal during Emotional Autobiographical Memory Recall

2013 ◽  
Vol 25 (1) ◽  
pp. 87-108 ◽  
Author(s):  
Alisha C. Holland ◽  
Elizabeth A. Kensinger
Keyword(s):  
Neural Activity ◽  
Visual Imagery ◽  
Time Course ◽  
Cognitive Reappraisal ◽  
Neural Correlates ◽  
Memory Recall ◽  
Emotional Intensity ◽  
Autobiographical Memories ◽  
Neural Processes ◽  
Emotion Generation

We used fMRI to investigate the neural processes engaged as individuals down- and up-regulated the emotions associated with negative autobiographical memories (AMs) using cognitive reappraisal strategies. Our analyses examined neural activity during three separate phases, as participants (a) viewed a reappraisal instruction (i.e., Decrease, Increase, Maintain), (b) searched for an AM referenced by a self-generated cue, and (c) elaborated upon the details of the AM being held in mind. Decreasing emotional intensity primarily engaged activity in regions previously implicated in cognitive control (e.g., dorsal and ventral lateral pFC), emotion generation and processing (e.g., amygdala, insula), and visual imagery (e.g., precuneus) as participants searched for and retrieved events. In contrast, increasing emotional intensity engaged similar regions during the instruction phase (i.e., before a memory cue was presented) and again as individuals later elaborated upon the details of the events they had recalled. These findings confirm that reappraisal can modulate neural activity during the recall of personally relevant events, although the time course of this modulation appears to depend on whether individuals are attempting to down- or up-regulate their emotions.

scholarly journals Multi-Timescale Drowsiness Characterization Based on a Video of a Driver’s Face

Sensors ◽  
2018 ◽  
Vol 18 (9) ◽  
pp. 2801 ◽  
Author(s):  
Quentin Massoz ◽  
Jacques Verly ◽  
Marc Van Droogenbroeck
Keyword(s):  
Real Time ◽  
Facial Expressions ◽  
Cross Validation ◽  
Time Window ◽  
Reaction Times ◽  
Ground Truth ◽  
Trade Off ◽  
Trade Offs ◽  
Psychomotor Vigilance ◽  
Long Timescales

Drowsiness is a major cause of fatal accidents, in particular in transportation. It is therefore crucial to develop automatic, real-time drowsiness characterization systems designed to issue accurate and timely warnings of drowsiness to the driver. In practice, the least intrusive, physiology-based approach is to remotely monitor, via cameras, facial expressions indicative of drowsiness such as slow and long eye closures. Since the system’s decisions are based upon facial expressions in a given time window, there exists a trade-off between accuracy (best achieved with long windows, i.e., at long timescales) and responsiveness (best achieved with short windows, i.e., at short timescales). To deal with this trade-off, we develop a multi-timescale drowsiness characterization system composed of four binary drowsiness classifiers operating at four distinct timescales (5 s, 15 s, 30 s, and 60 s) and trained jointly. We introduce a multi-timescale ground truth of drowsiness, based on the reaction times (RTs) performed during standard Psychomotor Vigilance Tasks (PVTs), that strategically enables our system to characterize drowsiness with diverse trade-offs between accuracy and responsiveness. We evaluated our system on 29 subjects via leave-one-subject-out cross-validation and obtained strong results, i.e., global accuracies of 70%, 85%, 89%, and 94% for the four classifiers operating at increasing timescales, respectively.

scholarly journals Molecular Cross-Validation for Single-Cell RNA-seq

2019 ◽  
Author(s):  
Joshua Batson ◽  
Loïc Royer ◽  
James Webber
Keyword(s):  
Single Cell ◽  
Cross Validation ◽  
Individual Cell ◽  
Principal Component ◽  
Ground Truth ◽  
Rna Seq ◽  
Optimal Parameters ◽  
Denoising Method ◽  
Data Driven Approach ◽  
Cell Data

Single-cell RNA sequencing enables researchers to study the gene expression of individual cells. However, in high-throughput methods the portrait of each individual cell is noisy, representing thousands of the hundreds of thousands of mRNA molecules originally present. While many methods for denoising single-cell data have been proposed, a principled procedure for selecting and calibrating the best method for a given dataset has been lacking. We present “molecular cross-validation,” a statistically principled and data-driven approach for estimating the accuracy of any denoising method without the need for ground-truth. We validate this approach for three denoising methods—principal component analysis, network diffusion, and a deep autoencoder—on a dataset of deeply-sequenced neurons. We show that molecular cross-validation correctly selects the optimal parameters for each method and identifies the best method for the dataset.

scholarly journals AI4CoV: Matching COVID-19 Patients to Treatment Options Using Artificial Intelligence

2020 ◽  
Author(s):  
Andrew I. Hsu ◽  
Amber S. Yeh ◽  
Shao-Lang Chen ◽  
Jerry J. Yeh ◽  
DongQing Lv ◽  
...  
Keyword(s):  
Clinical Trials ◽  
Language Processing ◽  
Cross Validation ◽  
Early Stage ◽  
Treatment Options ◽  
Clinical Manifestations ◽  
Ground Truth ◽  
English Text ◽  
Eligibility Criteria ◽  
Clinical Notes

AbstractWe developed AI4CoV, a novel AI system to match thousands of COVID-19 clinical trials to patients based on each patient’s eligibility to clinical trials in order to help physicians select treatment options for patients. AI4CoV leveraged Natural Language Processing (NLP) and Machine Learning to parse through eligibility criteria of trials and patients’ clinical manifestations in their clinical notes, both presented in English text, to accomplish 92.76% AUROC on a cross-validation test with 3,156 patient-trial pairs labeled with ground truth of suitability. Our retrospective multiple-site review shows that according to AI4CoV, severe patients of COVID-19 generally have less treatment options suitable for them than mild and moderate patients and that suitable and unsuitable treatment options are different for each patient. Our results show that the general approach of AI4CoV is useful during the early stage of a pandemic when the best treatments are still unknown.

scholarly journals Bimodality of Theta Phase Precession in Hippocampal Place Cells in Freely Running Rats

2002 ◽  
Vol 87 (6) ◽  
pp. 2629-2642 ◽  
Author(s):  
Yoko Yamaguchi ◽  
Yoshito Aota ◽  
Bruce L. McNaughton ◽  
Peter Lipa
Keyword(s):  
Place Cells ◽  
Distribution Functions ◽  
Data Sets ◽  
Place Field ◽  
Spike Density ◽  
Phase Versus ◽  
Normal Probability ◽  
Phase Precession ◽  
Theta Phase ◽  
Theta Phase Precession

The firing of hippocampal principal cells in freely running rats exhibits a progressive phase retardation as the animal passes through a cell's “place” field. This “phase precession” is more complex than a simple linear shift of phase with position. In the present paper, phase precession is quantitatively analyzed by fitting multiple (1–3) normal probability density functions to the phase versus position distribution of spikes in rats making repeated traversals of the place fields. The parameters were estimated by the Expectation Maximization method. Three data sets including CA1 and DG place cells were analyzed. Although the phase-position distributions vary among different cells and regions, this complexity is well described by a superposition of two normal distribution functions, suggesting that the firing behavior consists of two components. This conclusion is supported by the existence of two distinct maxima in the mean spike density in the phase versus position plane. In one component, firing phase shifts over a range of about 180°. The second component, which occurs near the end of the traversal of the place field, exhibits a low correlation between phase and position and is anti-phase with the phase-shift component. The functional implications of the two components are discussed with respect to their possible contribution to learning and memory mechanisms.

scholarly journals Initial Studies with the Lightning Detector on the C/NOFS Satellite, and Cross Validation with WWLLN

2011 ◽  
Vol 28 (11) ◽  
pp. 1423-1435 ◽  
Author(s):  
Abram R. Jacobson ◽  
Robert H. Holzworth ◽  
Michael P. McCarthy ◽  
Robert F. Pfaff
Keyword(s):  
High Power ◽  
Hot Spots ◽  
Cross Validation ◽  
Ground Truth ◽  
Field Of View ◽  
Congo Basin ◽  
Maritime Continent ◽  
Low Latitudes ◽  
Extended Field ◽  
Extended Field Of View

Abstract The lightning detector (LD) on the Communications/Navigation Outage Forecast System (C/NOFS) satellite uses a pair of silicon photodiodes, viewing each flank at right angles to the satellite track over an extended field of view. The data product is a report every ½ s of the number of digitizer cycles (125 μs each) for which the detected power was in predefined ranges. The performance of this system over the first 2.5 years of the C/NOFS mission is discussed, statistics of its lightning observations are presented, and a statistical cross validation using the World-Wide Lightning Location Network (WWLLN) as a ground truth is provided. It is found that the LD reports of lightning, despite their blunt timing (½ s), show correlation with strokes detected and located by WWLLN. The irradiance of these strokes lies on the high-power flank of the irradiance distribution seen earlier by the FORTE satellite. Thus, the LD thresholds favor high-power lightning; it is shown that the closest portion of the field of view is more likely to provide WWLLN coincidences than is the furthest portion of the field of view. Statistics of lightning incidence are examined at low latitudes, versus longitude, and distributions that are consistent with those established earlier by the OTD and LIS instruments are retrieved. Finally, the longitude dependence of the irradiance per stroke is examined and the ways in which it differs between the three major lightning “hot spots” is explored. It is observed that the radiance per stroke over the Congo Basin is lower than that over the other two hot spots (Maritime Continent/South Asia and the Americas), consistent with earlier observations by the OTD imager.

Sign in / Sign up

Export Citation Format

Share Document