We propose a deep convolutional neural network architecture codenamed Inception that achieves the new state of the art for classification and detection in the ImageNet Large-Scale Visual Recognition Challenge 2014 (ILSVRC14). The main hallmark of this architecture is the improved utilization of the computing resources inside the network. By a carefully crafted design, we increased the depth and width of the network while keeping the computational budget constant. To optimize quality, the architectural decisions were based on the Hebbian principle and the intuition of multi-scale processing. One particular incarnation used in our submission for ILSVRC14 is called GoogLeNet, a 22 layers deep network, the quality of which is assessed in the context of classification and detection.

/pdf/going-deeper-with-convolutions-1yobw2o2ds.pdf

Going deeper with convolutions

Co-Planar Stereotaxic Atlas of the Human Brain—3-Dimensional Proportional System: An Approach to Cerebral Imaging, J. Talairach, P. Tournoux. Georg Thieme Verlag, New York (1988), 122 pp., 130 figs. DM 268

I have developed "tennis elbow" from lugging this book around the past four weeks, but it is worth the pain, the effort, and the aspirin. It is also worth the (relatively speaking) bargain price. Including appendixes, this book contains 894 pages of text. The entire panorama of the neural sciences is surveyed and examined, and it is comprehensive in its scope, from genomes to social behaviors. The editors explicitly state that the book is designed as "an introductory text for students of biology, behavior, and medicine," but it is hard to imagine any audience, interested in any fragment of neuroscience at any level of sophistication, that would not enjoy this book. The editors have done a masterful job of weaving together the biologic, the behavioral, and the clinical sciences into a single tapestry in which everyone from the molecular biologist to the practicing psychiatrist can find and appreciate his or

Principles of Neural Science

http://www.cnbc.pt/jpmatos/29.%20Bradley.pdf?origin=publicationDetail

Measuring emotion: The self-assessment manikin and the semantic differential

Well-being : the foundations of hedonic psychology

Emotion and Motivation in the Perceptual Processing of Natural Scenes

https://csea.phhp.ufl.edu/Publications/322.%20Weymar_EncodingReinstateThreat_2014.pdf

Encoding and reinstatement of threat: Recognition potentials

There is abundant evidence from memory research using event-related potentials (ERPs) that items that have been experienced before elicit a larger late (500–800 ms) centro-parietal positive potential than new items when presented in the context of a recognition test (for reviews see Friedman & Johnson, 2000; Rugg & Curran, 2007) Whether stimuli are emotionally arousing or neutral, differences in ERP positivity between old and new items has been reliably found during explicit recognition for words (Windmann & Kutas, 2001), faces (Johansson et al, 2004), and scenes (Koenig and Mecklinger, 2008; Versace et al, 2010; Weymar et al, 2010) The amplitude of the difference, however, is often affected by whether the post-encoding task explicitly probes episodic memory or not (eg, Rugg & Wilding, 2000)

For instance, old-new ERP differences are attenuated when old and new words are presented in the context of a semantic categorization task (Duzel et al, 1999; Rugg & Wilding, 2000) Using natural scenes, a robust old/new ERP difference was found during explicit recognition, but when old and new exemplars were simply viewed with no task, a significant old-new ERP difference was found only for emotional scenes (Ferrari et al, 2012) Assuming the late old-new ERP difference reflects retrieval from episodic memory (Rugg & Curran, 2007), these data suggest that not all post-encoding tasks prompt episodic retrieval, and that emotional stimuli may be more likely to be spontaneously retrieved in the absence of explicit recognition

In two experiments, we further investigated the brain dynamics of implicit and explicit retrieval In Experiment 1, old and new pictures were presented in different instructional contexts following encoding: One group performed a semantic categorization task, in which the participant decided whether each picture included one person or more than one person and pressed a button indicating their decision This task, like explicit recognition (and unlike free viewing) requires both an active decision and a motor response Old-new ERPs measured during semantic categorization were compared to ERPs elicited in a group of participants who either made an explicit recognition decision (ie old/new) or simply viewed old and new pictures; both of these conditions are identical to those used by Ferrari et al (2012)

Differences in ERP timing and topography have previously been suggested to index different mnemonic processes (Rugg and Curran, 2007) An early (300–500 ms) difference that is maximal over frontal sensors is often attributed to decisions that are based on familiarity (Rugg & Curran, 2007), while a later (>500 ms) difference that is maximal over central sensors has been attributed to explicit recollection, as it is enhanced by depth of processing, correct source judgments (eg, Curran, Tepe, & Piatt, 2006; Rugg & Curran, 2007) and confidence ratings (Weymar et al, 2009) Previous ERP studies have reported similar old/new differences in the late ERP for both emotional and neutral pictures on an immediate recognition test (Koenig & Mecklinger, 2008; Ferrari et al, 2012; Versace et al, 2006, 2010), whereas a larger old/new difference for emotional, compared to neutral, pictures has been reported after longer retention intervals (eg, Schaefer et al, 2010; Weymar et al, 2009) In Experiment 1, we assessed effects of task context on both early and late components of the ERP

Of critical interest are ERP old/new differences for emotional and neutral pictures presented in the context of a semantic categorization (one or more people?) task, which does not explicitly probe episodic memory Because a number of studies have reported old-new differences in implicit (or indirect) tasks (eg, Guillem et al, 2000; Kazmerski & Friedman, 1997), one hypothesis is that an old-new ERP difference will be found for both emotional and neutral pictures during categorization If so, the previous null effect for neutral pictures during free viewing may simply reflect the lack of a requirement to process the picture On the other hand, if spontaneous retrieval is more likely for affectively arousing cues when memory retrieval is not explicitly required, enhanced ERP positivity for old, compared to new, pictures will again be found for emotional, but not neutral, pictures in the semantic categorization task, paralleling the effects reported by Ferrari et al (2012) during free viewing

https://csea.phhp.ufl.edu/Publications/318.%20Weymar_ExplicitSpontaneousRetrieval_2013.pdf

Explicit and spontaneous retrieval of emotional scenes: electrophysiological correlates.

. Several decades of research are reviewed, assessing patterns of psychophysiological reactivity in anxiety patients responding to a fear/threat imagery challenge. Findings show substantive differences in these measures within principal diagnostic categories, questioning the reliability and categorical specificity of current diagnostic systems. Following a new research framework (US National Institute of Mental Health [NIMH], Research Domain Criteria [RDoC]; Cuthbert & Insel, 2013), dimensional patterns of physiological reactivity are explored in a large sample of anxiety and mood disorder patients. Patients’ responses (e.g., startle reflex, heart rate) during fear/threat imagery varied significantly with higher questionnaire measured “negative affect,” stress history, and overall life dysfunction – bio-marking disorder groups, independent of Diagnostic and Statistical Manuals (DSM). The review concludes with a description of new research, currently underway, exploring brain function indices (stru...

The Psychophysiology of Anxiety and Mood Disorders

The climate in which scientific research is conducted changes over time, and in recent years there has been a shift from a positive view of programmatic science to a more negative evaluation that its contribution to scientific progress is only incremental. In this special issue focusing on the tools of a programmatic approach—replication, reliability and reproducibility—I reflect on changes in scientific practice over my research career, considering some factors contributing to changes in emphasis and highlighting potential pitfalls, particularly in terms of the impact on scientific progress and future scientists. In concluding, I suggest that, as members of the scientific community, we can influence current scientific practices in our day-to-day roles as authors, reviewers, investigators, editors, employers, and educators.

Margaret M. Bradley

Papers

Emotion and Motivation in the Perceptual Processing of Natural Scenes

Encoding and reinstatement of threat: Recognition potentials

Explicit and spontaneous retrieval of emotional scenes: electrophysiological correlates.

The Psychophysiology of Anxiety and Mood Disorders

The science pendulum: From programmatic to incremental-and back?