About
Activity
-
[CONCOURS] On est monté sur les planches … … et c’était pas du théâtre. 🎉 BavAR[t] a remporté le Coup de cœur du public lors du Concours de la…
[CONCOURS] On est monté sur les planches … … et c’était pas du théâtre. 🎉 BavAR[t] a remporté le Coup de cœur du public lors du Concours de la…
Liked by Rob Cao
-
We often witness a tendency to oversimplify complex issues by fixating on single elements. This attitude, while appealing in its simplicity, often…
We often witness a tendency to oversimplify complex issues by fixating on single elements. This attitude, while appealing in its simplicity, often…
Liked by Rob Cao
-
Excited to welcome my close collaborator and great friend, Kenneth Davies to the sunny shores of Singapore!🇸🇬☀️ Last week, we had a fantastic time…
Excited to welcome my close collaborator and great friend, Kenneth Davies to the sunny shores of Singapore!🇸🇬☀️ Last week, we had a fantastic time…
Liked by Rob Cao
Experience & Education
Publications
-
Binocular rivalry reveals an out-of-equilibrium neural dynamics suited for decision-making
eLife
In ambiguous or conflicting sensory situations, perception is often ‘multistable’ in that it perpetually changes at irregular intervals, shifting abruptly between distinct alternatives. The interval statistics of these alternations exhibits quasi-universal characteristics, suggesting a general mechanism. Using binocular rivalry, we show that many aspects of this perceptual dynamics are reproduced by a hierarchical model operating out of equilibrium. The constitutive elements of this model…
In ambiguous or conflicting sensory situations, perception is often ‘multistable’ in that it perpetually changes at irregular intervals, shifting abruptly between distinct alternatives. The interval statistics of these alternations exhibits quasi-universal characteristics, suggesting a general mechanism. Using binocular rivalry, we show that many aspects of this perceptual dynamics are reproduced by a hierarchical model operating out of equilibrium. The constitutive elements of this model idealize the metastability of cortical networks. Independent elements accumulate visual evidence at one level, while groups of coupled elements compete for dominance at another level. As soon as one group dominates perception, feedback inhibition suppresses supporting evidence. Previously unreported features in the serial dependencies of perceptual alternations compellingly corroborate this mechanism. Moreover, the proposed out-of-equilibrium dynamics satisfies normative constraints of continuous decision-making. Thus, multistable perception may reflect decision-making in a volatile world: integrating evidence over space and time, choosing categorically between hypotheses, while concurrently evaluating alternatives.
Other authorsSee publication -
Excitatory and inhibitory subnetworks are equally selective during decision-making and emerge simultaneously during learning
Neuron - Cell Press
Inhibitory neurons, which play a critical role in decision-making models, are often simplified as a single pool of non-selective neurons lacking connection specificity. This assumption is supported by observations in the primary visual cortex: inhibitory neurons are broadly tuned in vivo and show non-specific connectivity in slice. The selectivity of excitatory and inhibitory neurons within decision circuits and, hence, the validity of decision-making models are unknown. We simultaneously…
Inhibitory neurons, which play a critical role in decision-making models, are often simplified as a single pool of non-selective neurons lacking connection specificity. This assumption is supported by observations in the primary visual cortex: inhibitory neurons are broadly tuned in vivo and show non-specific connectivity in slice. The selectivity of excitatory and inhibitory neurons within decision circuits and, hence, the validity of decision-making models are unknown. We simultaneously measured excitatory and inhibitory neurons in the posterior parietal cortex of mice judging multisensory stimuli. Surprisingly, excitatory and inhibitory neurons were equally selective for the animal’s choice, both at the single-cell and population level. Further, both cell types exhibited similar changes in selectivity and temporal dynamics during learning, paralleling behavioral improvements. These observations, combined with modeling, argue against circuit architectures assuming non-selective inhibitory neurons. Instead, they argue for selective subnetworks of inhibitory and excitatory neurons that are shaped by experience to support expert decision-making.
-
Avoiding degradation in deep feed-forward networks by phasing out skip-connections
International Conference on Artificial Neural Networks
A widely observed phenomenon in deep learning is the degradation problem: increasing the depth of a network leads to a decrease in performance on both test and training data. Novel architectures such as ResNets and Highway networks have addressed this issue by introducing various flavors of skip-connections or gating mechanisms. However, the degradation problem persists in the context of plain feed-forward networks. In this work we propose a simple method to address this issue. The proposed…
A widely observed phenomenon in deep learning is the degradation problem: increasing the depth of a network leads to a decrease in performance on both test and training data. Novel architectures such as ResNets and Highway networks have addressed this issue by introducing various flavors of skip-connections or gating mechanisms. However, the degradation problem persists in the context of plain feed-forward networks. In this work we propose a simple method to address this issue. The proposed method poses the learning of weights in deep networks as a constrained optimization problem where the presence of skip-connections is penalized by Lagrange multipliers. This allows for skip-connections to be introduced during the early stages of training and subsequently phased out in a principled manner. We demonstrate the benefits of such an approach with experiments on MNIST, fashion-MNIST, CIFAR-10 and CIFAR-100 where the proposed method is shown to greatly decrease the degradation effect (compared to plain networks) and is often competitive with ResNets.
Other authorsSee publication -
Collective Activity of Many Bistable Assemblies Reproduces Characteristic Dynamics of Multistable Perception
The Journal of Neuroscience
The timing of perceptual decisions depends on both deterministic and stochastic factors, as the gradual accumulation of sensory evidence (deterministic) is contaminated by sensory and/or internal noise (stochastic). When human observers view multistable visual displays, successive episodes of stochastic accumulation culminate in repeated reversals of visual appearance. Treating reversal timing as a “first-passage time” problem, we ask how the observed timing densities constrain the underlying…
The timing of perceptual decisions depends on both deterministic and stochastic factors, as the gradual accumulation of sensory evidence (deterministic) is contaminated by sensory and/or internal noise (stochastic). When human observers view multistable visual displays, successive episodes of stochastic accumulation culminate in repeated reversals of visual appearance. Treating reversal timing as a “first-passage time” problem, we ask how the observed timing densities constrain the underlying stochastic accumulation. Importantly, mean reversal times (i.e., deterministic factors) differ enormously between displays/observers/stimulation levels, whereas the variance and skewness of reversal times (i.e., stochastic factors) keep characteristic proportions of the mean. What sort of stochastic process could reproduce this highly consistent “scaling property?” Here we show that the collective activity of a finite population of bistable units (i.e., a generalized Ehrenfest process) quantitatively reproduces all aspects of the scaling property of multistable phenomena, in contrast to other processes under consideration (Poisson, Wiener, or Ornstein-Uhlenbeck process). The postulated units express the spontaneous dynamics of attractor assemblies transitioning between distinct activity states. Plausible candidates are cortical columns, or clusters of columns, as they are preferentially connected and spontaneously explore a restricted repertoire of activity states. Our findings suggests that perceptual representations are granular, probabilistic, and operate far from equilibrium, thereby offering a suitable substrate for statistical inference.
Other authorsSee publication -
Stochastic Accumulation by Cortical Columns May Explain the Scalar Property of Multistable Perception
Physical Review Letters -
The timing of certain mental events is thought to reflect random walks performed by underlying neural dynamics. One class of such events—stochastic reversals of multistable perceptions—exhibits a unique scalar property: even though timing densities vary widely, higher moments stay in particular proportions to the mean. We show that stochastic accumulation of activity in a finite number of idealized cortical columns—realizing a generalized Ehrenfest urn model—may explain these observations…
The timing of certain mental events is thought to reflect random walks performed by underlying neural dynamics. One class of such events—stochastic reversals of multistable perceptions—exhibits a unique scalar property: even though timing densities vary widely, higher moments stay in particular proportions to the mean. We show that stochastic accumulation of activity in a finite number of idealized cortical columns—realizing a generalized Ehrenfest urn model—may explain these observations. Modeling stochastic reversals as the first-passage time of a threshold number of active columns, we obtain higher moments of the first-passage time density. We derive analytical expressions for noninteracting columns and generalize the results to interacting columns in simulations. The scalar property of multistable perception is reproduced by a dynamic regime with a fixed, low threshold, in which the activation of a few additional columns suffices for a reversal.
Other authorsSee publication
Languages
-
Français
Native or bilingual proficiency
-
English
Native or bilingual proficiency
-
Italiano
Professional working proficiency
-
Deutsch
Limited working proficiency
-
Portuguese
Elementary proficiency
More activity by Rob
-
Getting on stage in front of a lot of people isn't easy, but the presenters at the Spatial Data Science Conference made it look that way. I want to…
Getting on stage in front of a lot of people isn't easy, but the presenters at the Spatial Data Science Conference made it look that way. I want to…
Liked by Rob Cao
-
Started a new role last week with the Data Science team at choreograph led by Ngan Meng Tan. Had the pleasure to meet the entire team today spreading…
Started a new role last week with the Data Science team at choreograph led by Ngan Meng Tan. Had the pleasure to meet the entire team today spreading…
Liked by Rob Cao
-
Thanks SafeGraph for inviting me to present with Jaime Unger at CARTO’s Spatial Data Science Conference about how Mobsta Ltd - Certified B Corp®…
Thanks SafeGraph for inviting me to present with Jaime Unger at CARTO’s Spatial Data Science Conference about how Mobsta Ltd - Certified B Corp®…
Liked by Rob Cao
-
#SDSC24 My first Spatial Data Science Conference in London and it was AMAZING! Many thanks to our customers and partners. Thank you to the…
#SDSC24 My first Spatial Data Science Conference in London and it was AMAZING! Many thanks to our customers and partners. Thank you to the…
Liked by Rob Cao
-
I had a great time talking with Robert Clements and Cody about Generative AI, Roblox, and research. You can listen to our discussion on the USF Data…
I had a great time talking with Robert Clements and Cody about Generative AI, Roblox, and research. You can listen to our discussion on the USF Data…
Liked by Rob Cao
-
Great Spatial Data Science conference yesterday at the Royal Geographic Society in London. Through the presentation and discussion with peers, we…
Great Spatial Data Science conference yesterday at the Royal Geographic Society in London. Through the presentation and discussion with peers, we…
Liked by Rob Cao
-
Had an awesome time yesterday at CARTO's Spatial Data Science Conference, sharing the stage with Hannah Babineau and Paul Langston for the panel…
Had an awesome time yesterday at CARTO's Spatial Data Science Conference, sharing the stage with Hannah Babineau and Paul Langston for the panel…
Shared by Rob Cao
-
From Flood Analytics 🌧 to Digital Footprints 👣 at #SDSC24! 🌧️ 🌊 Did you know that flooding costs the UK £2.2bn annually, affecting…
From Flood Analytics 🌧 to Digital Footprints 👣 at #SDSC24! 🌧️ 🌊 Did you know that flooding costs the UK £2.2bn annually, affecting…
Liked by Rob Cao
-
🇬🇧 Attending #SDSC24 next week? 👋 Swing by our stand and meet a few faces behind Data Appeal! Chat with Rudy Becarelli, Ph.D., Marco Ciabatti…
🇬🇧 Attending #SDSC24 next week? 👋 Swing by our stand and meet a few faces behind Data Appeal! Chat with Rudy Becarelli, Ph.D., Marco Ciabatti…
Liked by Rob Cao
People also viewed
Explore collaborative articles
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore MoreOthers named Rob Cao
-
Roberto (Rob) Cao
Host of The Flame: Entertainment & Sports
-
Robert CAO
-
Robert Cao
Values driven media, marketing and communications - inspiring good people to dream big, share their story and achieve their purpose. Be kind, do the right thing.
-
Robert Cao
54 others named Rob Cao are on LinkedIn
See others named Rob Cao