Friday, July 12, 2019

Myths of bio-inspired deep neural networks.


In my opinion neuroscience best aid DNN design if the designers are having stability and generalizability in mind, instead of accuracy and speed. brains are evolved to be a fault-tolerant, failure-proof, multi-tasking system, where most functions can be achieved via multiple mechanisms, and there are “backup engines” to prevent system crash. Nervous system is also a analog wetware where the algorithms has to be implement with different logic from computer algorithms. Besides, there are also infinite amount of chemical-biology signaling and modulating pathways that run in tandem with all other organ systems to support basic metabolism and immune response. Radically speaking, you can survive without halve a brain, but not without half your heart. Reasonably, the existence of brain in not to master a single functions such as image segmentation, but make it just good enough to get the final reward ‘to survive’.  

Thursday, December 6, 2018

Book Review : How to create a mind , Ray Kruzweil


The author Ray Kruzweil is known for his audacious prediction about near future. His prediction usually comes from his empirical understanding of how things work the way Greeks philosophers do.  He extrapolates his theories derived from 'thought experiments' to the maximum, and regress them until it fits empiricality. He has written many books that predicts future, and one of my favorite is 'singularity is near'. That book concludes that artificial reality is going to outsmart human at exponential rate. The prediction appears to be right in retrospect. Recent news from Google shows that NLP is already good enough to fool real person, and object recognition has out-perform humans in some competition. Some of those once considered the most difficult question a decade ago now looks basic to most engineers. The rate of improvement in AI indeed accelerates exponentially. This law is coined by Kruzweil as LOAR( LAW OF ACCELERATING RETURN)

He claims all information technology follows the same fashion, and Moore's Law incidentally being the most successful one. After all, our modern world is built on the computing power per unit cost, which essentially relies on miniaturization of transistors that improves at exponential rate.

The book How To Create a Mind tries to address a even more complicated question that humans have pondered since the start of history. How does mind work ? Or from engineering perspective,  how to create a mind ?

The book leads us through a series of thought experiments to deduce some basic properties of psychophysics. Some of the properties are already known by philosophers for a long time, but the author put them in light of artificial intelligence.

Chapter 1 :

Our mind that is able to retrospect the existence of itself is a unique ability of human being. Descartes's 'I think, therefore I am' can by self-evident, but it never lands on solid scientific ground. Consciousness, free will, and mind are just alternative words for soul in religious era. The essence of it is still elusive, and that's why having a theory of min is important. evidences of neuroscience are like words, but need a language -- a theory to put them into sentence. We need a unified model of our minds in order to understand how minds work.

However, is it achievable ?  The author believe it is achievable based on how scientists have successfully discovered Theory of Evolution and Theory of Relativity. They reminds us most people are limited by their ambition to see beyond their peers' believes.

Chapter 2:

The book is gigantic and worth reading times and again so here I just excerpt some quotation as footnotes of the chapter.
Q : Does the brain work the way how computers work ? 
No. We don't store every details of episodes of life in our brains. Our memories are stored in hierarchical manner. We only perceive and store those that are important to us. The selective filtering is achieved by a hierarchically organized receptive fields. Each layers reduce the dimensions (complexity) of features to push the neural networks to learn the representations is the most efficient way by attempting to minimize prediction errors. In retrospect, such hierarchical representation is effective because our world consists of objects that can be modularized into attributes of hierarchical abstractions. We recall our memories by reconstructing the input using the stored coding of the neuronal network through hierarchical layers(inverse forward method). This explains why our memories can be faulty in details while the emotional tags remain accurate.  Not only our explicit memories are stored in hierarchical patterns. The procedural routines of actions are stored in hierarchical patterns. Therefore we are able to generalize our motor controls by reusing the patterns. The same type of hierarchy also presents in our ability to recognize contexts and situations.   

 Our memories are stored in sequential patterns. We can't recall a series of number in reverse order easily. However,  they can be easily accessed in the order they are remembered. We are unable to directly reverse the sequence of a memory. This is manifested by the fact that associative learning of declarative memory (episodic and semantic) are encoded with LTP and LTD of inter-synaptic connections. The presynaptic terminal release neurotransmitters and post-synaptic dendrites have NT receptors to receive them. So the signaling pathways are directional.
We can recognize a pattern even if only part of it is perceived (seen, heard, felt), and even if it contains alterations.
Our recognition ability is apparently able to detect invariant features of a pattern -- characteristics that survive real-world variations.
Our conscious experience of our perceptions is actually changed by our interpretations. To solve a certain problem we need a certain set of policies to plan. However, the problem in real world is ambiguously defined. Therefore different interpretations occurs because each individual has varying policies and priors. We are constantly predicting the future and hypothesizing what we will experience. This expectation influences what we actually perceive.

Chapter 3 :
Neocortex is pattern recognition machine.

One Page Review of Recent Improvement of Deep Neural Networks in Optical Object Recognition and Detection.



Background :


In spite of tremendous success of neural networks in virtually all fronts of machine learning battle ground since the early 2000s, let us not forget it was not so long ago a 20 years long winter of the field since it being sentenced dead by Marvin Minsky, after his decisive book Perceptrons got published in 1969. In this monumental book, Minsky mathematically showed that the idea of perceptrons, proposed by Frank Rosenblat in 1957, although promising at some domains, cannot to perform XOR operation. Despite the claim is only true on linear single layer perceptrons, Minsky’s preeminent status in AI at the time led to significant decline in interests and fundings in research of neural networks for the last 20 years until around 2000.  

Architectures :


LeNet

Second wave of popularity of neural networks can be dates back to the success of LeNet-5 in solving MNIST handwritten digit recognition. Yann LeCun in his review paper in 1998 claimed that gradient based learning algorithm coupled with appropriated architecture can deal with image recognition with efficiency and accuracy. Convolutional Neural networks, which in essence utilizes strong local correlation in 2D space,  as well as weights sharing and sub-sampling to deal shift-invariance as a feature extracting scheme for images, and classify the feature vectors with multi-layer perceptrons lays the foundation of renaissance of neural networks.

AlexNet

In 2012, Alex Krizhevsky’s AlexNet set new record for ILSVRC classification challenge with top-5 error of only 15.3%. The second best model’s top-5 error is 26.2%. It consists five convolutional layers alternate with max-pooling layers, joined with 2 fully connected layers. AlexNet proved wider networks with sufficient training data can significantly outperform non-gradient based classifiers on image classification. Several techniques are introduced to achieve such improvement, such as using 2 GPUs, data augmentation, relu activation, and drop-out regularization.   

VGG

Karen Simonyan and Andrew Zissermanʻs VGG net is the 2nd runner in 2014 ILSVRC. Its architecture is very similar to AlexNet and LeNet, but with only one pipeline thanks to the improvement of GPUs. The network consists of 5 feature extraction blocks. Each of the feature extraction blocks consists of multiple convolution layers, and ends with one max-pooling layer. The feature extraction module is followed by a classification modules consists of three  fully connected layers. The advantage of VGG is probably its simple-to-implement structure, and open-sourced pretrained weights, making it easy to understand and deploy for various goals. However, it suffers from massive amount of parameters adding up to 550 MB, therefore the training from scratch is slow.

GoogLeNet

Champion of 2014 ILSVRC developed by Google. As we can find VGG cannot go deeper in number of layers because the accuracy drops as it go deeper with that kind of architecture (called degradatio problem). GoogLeNet first introduce the idea of ‘inception’, by breaking apart fix sized convolutional kernel into multiple convolutional modules with various depths and kernel sizes. It helps extracts feature better and easier because hand-picked fixed sized kernel is not optimal for feature extraction, and the size is not trainable. With multiple feature extractors of different designs, we avoid this inductive bias in architecture design. Besides, inceptions effectively creates an ensemble of modules of different inductive biases and therefore combining models can find better features. As a result, it generalize better and become easier to train. It also introduce the bottle net 1x1 convolution, and use average pooling to replace FC layers to reduce parameters. To conclude, fractionalization allows architecture to go deeper. GoogLeNet comes with several upgrade versions combining tricks from other networks such as residual learning, batch normalization.

RESNET

As demonstrated by GoogLeNet, going deeper can improve accuracy if 1) right architecture, 2) sufficient gradient. Resnet allows training deeper network as large as 152 layers. The contribution of the network is they found identity mapping for standard convolutional or dense layer block is not trivial, therefore the residual connection from input to output of one standard convolutional block solve the problem. When identity mapping is optimal, model can simply set the weight of residual weight to 1, while the standard blocks learns residual correlations of data. It is easy to understand in the beginning of training, the gradient vanishes across layers. Residual connection ‘high way’ allows signals to go from the last layer to the first few layers.

DENSNET

DenseNet can be seen as a variation of resnet. It put residual connections not only between adjoint blocks, but between every other blocks. It further improves the training efficiency. Besides, the classifier uses features of multiple blocks, not just the last blocks. The increase of complexity of this kind of ‘skip connections’ of model can be seen as a way of ensemble learning.

Algorithms :


Rectified Linear Unit (ReLU) activation function is used to solve the problem of vanishing gradients resulting from the saturation of output passing through the sigmoid function. ReLU allows training deeper network( number of layers larger than 8) effectively without having to normalize input to activation function.  

Local response normalization (LRN) contrasts the local difference in the feature map, while normalize local response if they are relatively homogeneous. The effect is equivalent to contrasts high frequency difference, while damps low frequency difference.  

Overlapping Max Pooling is a sub-sampling techniques that replaces average pooling of LeNet-5, and also outdo non-overlapping max pooling. The author claims overlapping helps prevent loss of information and therefore reduce error.

Drop-Out is a technique to prevent overfitting by arbitrarily deactivate neurons with a fixed probability to prevent networks overly rely on a subset of active neurons. It forces all subsets of neurons to have equal chance to learn in a stochastic fashion. The benefit is threefold. It prevents some neurons from dying, so the learning is more distributed and less likely to fall into local minimum. Second,it effectively reduce biases the same fashion as ensemble learning. Finally, it mitigates bias of initial weights by ‘shuffling’ the model ensembles to train for each training iteration.     

Xavier Initialization is a technique to prevent the bad initialization of weights. Bad initialization could lead to dead neurons and weak gradients when the biases and weights saturate activation functions( either the ends of sigmoid, or the negative side of ReLu). It is important in training very deep networks because gradients and activation tends to attenuates over layers. It also demonstrates good initialization does matter.

Batch Normalization. The author argues standard optimization suffers from internal covariate shift. Normalize activations, to remove mean and unitize variance, prevents activation from saturations thanks to the non-linear activation functions, which leads to insensitive of neurons and vanishing gradients. The technique is more distinct especially when the network is deep.

ADAM optimizer is the standard optimizer whatever model you are training. It includes all good properties that has been demonstrated powerful in previous optimizers, such as  adaptive learning rate, exponential annealing, rescale the gradient so that the gradient stays invariant to re-scaling of objective function. Besides, it has a good property to prefer flat minimum basin.














Sunday, December 10, 2017

Excerpt form 'The Mind at Night'

The Mind at Night

Neuromodulators for wakefulness.

When we are awake, tow neuromodulators are essential to alert, waking consciousness : nerepinephrine ( help direct and focus our attention), and serotonin.Serotonin is not only in charge of mood regulation, but also judgment, learning, and memory. After we fall asleep, acetylcholine  excites visual, motor, and emotional centers of the brain and transmit signals that trigger rapid eye movements and visual imagery in dreams. (Ach is also responsible for neuro-muscluar junctions transmission in mammal.)


Anti Freudism
Hobson and McCarley's study maintained that since the signal that initiated the creation of dream imagery and came from brain stem and neocortex were just passively response. The resulting dream was the product of the forebrain making the best of the bad job producing even partially coherent dream imagery.

We forgot most of our dream because the two neuromodulators need to imprint them on memory are in short supply until we akwen.

PET scanning shows that prefrontal cortex takes the first and deepest plunge before the overall decrease of global activities. The area is the first to fall asleep and the last to come back on-line. Deactivation of the area is accompanied by sharp drops in levels of serotonin and nerepinephrine, which help us focus when we are awake. Then surge the acetylcholine (which fosters free-wheeling association) that turns on REM sleep. Its inactivity would explain why we have no reality testing in dreams.

REM may provide a situation where long-term memory traces can be processed off-line, either consolidated or pruned (LTP and LTD), while the prefrontal area is not actively processing information generated during REM itself.

Deja vu : 
The feeling that we have seen something before but we connot recall. Deja vu is like the episodic memory version of getting lost in a maze because of its grid cell firing pattern overlaps with another location in mental spatial map.



Hypothesis I  :  Clear address to empty house
Would it be possible that we dreamed of it before? Although our brain is designed not to remember the dream content (because it makes no sense in reality), but there's chances that some part of the memory circuits that were strengthened synthetically during sleep is activated by the daytime experience. This pathway is distinctive, but it doesn't contain interpretable information.

Hypothesis II : Associative emotional memory
Deja vu might be caused by activation of 'emotional memories'. Emotional memories are not declarable, but participate in the formation of declarable episodic memory. Deja vu may involves emotional memory, which also account for feeling of nostalgia. Nostalgia get summoned by songs and odor but not vision because Auditory pathway or olfactory pathway are more ancient and are therefore less filtered by our consciousness.  Their quick access to emotional memory reflects their nature of 'reptilian of brain'. The function of emotional memory in reptiles is directly relating to fight or flight behavior, where memory of fear and satisfaction decides the reward function of decision making.
The emotional tone of a current experience may evoke certain emotional memory due to their similarity in certain properties. In short, these two emotions associates to each other in the memory circuit. 

Hypothesis III : glitch of neural circuits
Deja vu may arise from spontaneous large scale synchronization (analogous to avalanche in complex system ) of association memory between current experience and a set of incomplete memories.

P.52 Maquet concluded that patterns of acturation in the amygdala and other cortical area provided a biological basis for the processing of memory.

P.77 MIT professor Matthew Wilson on mice dream and maze. Mice replay their day time experience of walking in a maze during sleep, according to the sequential firing pattern of place cells. This sequential memory replay may be the origin of latter evolved episodic memory of more intelligent primates.

REM sleep is crucial to consolidate procedual memory by by skiill rehersal, givien us an edge in survivial of the fittest.

P.84 If our emotions are aroused during an experience, our memory of it will be strengthened by that emotional tagging. Extreme emotional arousal, particularly stress increases the concentration of cortisol.


Ultimate Spin Doctor
P.122 
Thanks to brain-imaging studies showing the dreaming brain in action, the limbic system(amygdala, hippocampus,hypothalamus, mammilian body, cingulate gyrus,  entohrinal cortex, etc.), command center for directing emotion and storing strong emotional memory - is directing the dream show. Clearly, a dream is a product of multiple layers of mental activity. The brain is running tests of genetically programmed, survival-related behavior, as well as reviewing recent experience and integrating significant bits of new information with what's already in the memory database in order to update our individual concepts of how the world works and how we fit into it (reinforcement learning.) And having the brain's emotional center in the driver's seat means that the memories that are being singled out most prominently for processing are those that are emotionally charged : anxieties, feelings, of loss, blows to self-esteem, and physical or psychological trauma.  Meanwhile, all of this is taking place under unusual physiological conditions, in which the neuromodulators that help focus attention are in short supply and the brain responds only to its own internally generated signals. --This theory nicely explains why children are more prone to have nightmare than adults -- Their reptilian brain have larger influence on their dream due to late maturation of neocortex.

Brain can create pattern our of any noise (think of brain as as hierarchical pattern completion machine) The brain relentlessly try to create meaning out of whatever it encounters. The brain makes the best sense it can of the barrage of internally generated neural signals it receives, including random neural noise, and pass it to the next part of the system. If a visual image of two dots is created in the visual cortext, the parietal cortex is likely to turn them into a couple of eyes, and then put those two eyes into a face. If that face isn't recognized, the limbic system says 'Ah, that's worrisome; Let's run'. That signal goes to the motor system and the dream plot takes off. explains Antrobus.




Consciousness
According to Micheal Gazaniga's experiment with split-brain patients, the left hemisphere contains a neural system that he labels the 'interpreter', which continually seek to provide explanations for both internal and external perceptions and experiences.'Once you realize the brain is so gullible, you don't want to believe a damn thing it does. It alway tyring to make sense, and in doing so, it fabricates more than just dreams.'  says John Antobus. 'When we encounter a new decision, the emotional brain help us sort out which cognitive strategy to use'

Split patients spin stories, making up reasons to explain our actions. Maybe we make up the concept of consciousness to explain the unity of our self-identity when our attentions drift from one to another things constantly every second.

P.126 'It is the glue that unifies our story and creates our sense of being a whole, rational agent.'

P.183 "Why then should consciousness --whether in waking or dreaming form-- have emerged at all ? It may be because consciousness allows the system to plan future actions, opening up a potentially infinite behavioral repertoire and making explicit memory possible." says Koch. "Consciousness could involve synchronized firing of neurons at the millisecond level, whereas uncorrleated firing can influence behaviors without generating that special buzz in the head that makes us subjectively aware "


Vision is not just a bottom-up process, but also a top-down selection. We can recognize a character by seeing a fraction of it in a unconscious fashion. This pattern completion property is not presented in deep-learning. 


 P.144 Perhaps dreaming generates a wide range of behavioral schema. universal dreams are the equivalent of literary archetypes, akin to stories of jealousy, desire, and revenge that have unfolded from Greek mythology.

P.126 While all animals can learn to avoid food that makes them ill, only humans have the ability to ask why the plant made them sick. This ability to reason give human species a survival edge. Reasoning enable us to generalize the fragmented facts into knowledge. The same capacity to spin narratives in awake state is used to construct the tales that comprise our dreams, and eventually lead us to become psychologically interesting to ourselves as a species.

Epilogue
As the human brain evolved --acquiring the capability for language and for subjective awareness of our emotions.  -- dreaming took on added dimensions that reflected the growing complexity of our won brand of consciousness. As neural circuitry matures and becomes more sophisticated, human dreams ultimately become narratives, with the dreamer as the key player in a true-as-life reality that our brain constructs anew each night, drawing on both recent and long-term memory. Emotionally tagged memory has everything to do with our sense of self.



Wednesday, November 1, 2017

The Silicon Brain

The Silicon Brain
Author : P.S. Huang
May 25, 2014
                      July 5, 2048
  Hugo Nuland woke up fuzzy with every part of his body ached like burning.  He tried to recall what had just happened to him, but his brain hurts like there’s an hive in it, and the wasp herd went berserk whenever he tried to get focused. He is not the kind of person who get depressed easily, but this time he felt all his energy is drained in a way he’d never experienced before. For the first time in his 80 years’ legendary careers of multiple professions, he sensed a hint of terror residing in the dark corner in him. He tired to move his burning limbs, which responded him with severe cramps. He couldn’t help groaning out, only bitterly finding his throat too weak and too dry to utter a sound. He thought challengedly that he would die from fatigue, pain, and thirst in a minute, and he hated this arrangement of destiny not because of fear of death, his has indomitable courage, but he had a world changing secret needed to be told. Despite his strong motive of survival, he blacked out.    

    If there were not the ‘beep’ sounds hovering in the background constantly, one could even hear a needle drop off on the polished mozaic tiled floor. The place is freezing as if refrigerated that white, thick fog forms from people’s breath. A faint spotlight shed right from the centre of the dome, on to the centre of the room, where a egg-shaped capsule lays. Edges of the room is indiscernible. Darkness engulf everything.
     ‘Doctor, Nuland woke up.’  says a female voice apathetically. She is looking straight to the front, where nothing but other ‘mind workers’ seating in couches. In fact, they all look exactly identical with her. Every one got head shaved with finger-thick cords connect to back of their necks. All of them looks similar— pale skin and pale emotion. 
   Countless diagrams and curves are jumping in a small monitor embedded in the egg-shaped semitransparent capsule lying in the centre of the concentric circles of couches.  
  ‘ How is his sync value ?’ asks the deep, stern voice sent out from speaker , after a  short peak of electrical noise. Contrasting to the flat female voice, the man’s voice contains strong vigour. The voice echoes in the dark hall, then rippled away to the abyss of shear void.
  ‘99.9% Synced, Doctor. No special anomaly observed. We are at the finishing line. Congratulation, Doc. ’ The female voice rises again. It would be a sweet, cheering voice, were it not for its absence of soul. 
  ‘ Prepare for the unmount of the chip in 30 seconds, S.’  The man is calling the female, whose previous name is Sandra Pfizer. 
  ’ 16, 15, 14.’ As S counted, numerous black-suited people are watching all of this from a height.  Behind a curved glass shield, the unnoticeable space provides an hawk-eye view that snipes every moves under the dome.
‘ Here,’ the same deep, stern voices now sounds somewhat softened.
‘ Is where we call “Dome”, the very central kernel of the whole project.’ The ‘doctor’ proudly announces to his secret guests. ‘ Its capability will going to save millions of people from suffering both mental and physical illness.’ ‘ Handicap, paralysis, and muscle degeneration can be cured COMPLETELY.’ Doctor tries to induce interests of his private guests with this opening. ‘What is more, this technology will be a gospel to patients of mental illness. There will be no more people suffering depression, Parkinson’s, and Alzheimer’s disease.’  His bold claim seems to seize some curiosity of his audience, and as expected, he also notice some people stare at him in disbelief. 
‘Should you feel unconvinced of what I said, gentlemen.  May I invite you to have a short journey with me,’  When he voice ceases, he turn around to the glass shield, ‘Shall we ?’ he smiles, looking down everything beneath his feet.
‘3,2,1, zero.’ S counted with serene voice, like a responsible timer. ‘This will definitely sweep their mind.’ Doctor mutters, as he adroitly put on his gloves as rehearsed. 
‘How about you, G?’  Doctor’s deep voice rises again after a short pause.  G, a stout bald guy subconsciously nods his head, making the cord adjoint to his neck sways like a jumping rope.   
 ‘G, show his life index on the main screen. Tell me you’ve started scanning EET and NIRS already.’ 
‘ You don’t expect me to wait for you and your dumbass crew, do you?’  Doctor growls as he hurriedly leans forward onto the front penal which showed up from nowhere on the glass windshield, separating them form the refrigerated, bacteria free dome.

’ Scanning is already on the way, doctor.’  G, the little figure in a white robe suddenly speaks, making some shocked guests jumped from their toes. This incident make some other guests giggles. A few people glare at them indignantly.  

‘You’d better.’ The Doctor replied indifferently.  ’You don’t screw up this time, G.’ ‘ Aren’t they all robots ? Doctor ?’  One of the guests asks suspiciously, ‘How can it possibly err? ’ ‘ In some way, they preserved some human quality,’ Doctor explains carelessly, for he knows he has to leave his concentration to the big show. ‘It helps them perform better, believe it or not.’  

Next second, with a sharp beep, an animated 3D brain image around a football size is brought to their view, with some spots of the image are glowing, while the brightness fluctuates with rhythm. The blue coral reefs like image is so vivid, that sometimes people doubt if it is an illusion. ‘ Ultra Definition Holographic Interactive System,UDHIS’ Doctor explained with humble smile, ‘Another invention of our institute. It exploits an sophisticate instrument coverages two orthogonal polarised light of the same objects, in short.’ He speaks while pushing the “brain” away, right towards a guest in the crowd. Too late to dodge, the brain passes though a guests’s chest, and shows up from the other side.  ‘Holy God! ’ Cry out the startled guest, while all the other guests draw a step back from the brain. ‘Don’t be afraid, It won’t bite.’ Doctor smiles at his panicking audience amusingly, he never get tired of this trick.
 ‘Watch carefully, gentlemen.’ Doctor lowers his voice like a  game show host. ‘THIS is the legendary Hugo Nuland’s brain, gentlemen.’ He try to make himself not sound too excited. ‘We captured him just last month.’ 
He nods his head affirmatively to the guests, who are the delegates of  Peace-Prosper Committee of the Global Union. ‘What? How dare you!’ The dark-skinned Indian delegate protests. ‘His rights are protected by the Constitution of the Earth! It’s a criminal act!’ The former Minister of Humanity of Southern Asia cried out shockingly, several of his colleges nodded with agreement.   
  ‘ Mr. Sunjishran, my esteemed friend. ’ Doctor interrupts the gentleman from his endless complaints. ‘ I respect your opinion, and I admire your noble integrity,’ Doctor speaks mildly, holding his hands before his chest. ‘ But Mr. Nuland is not a human.’ Doctor’s words are almost a sigh but it can not be clearer. ‘ I will prove it to you.’ 
  He holds up a crown-like headset assembled with gadgets, ready to commence the surgery.      
   The coral of brain shimmers a splendid rainbow of sapphire blue, its valleys are darker than deepest water of Pacific Ocean. Streams of lights bustling around the network of bright spots, like traffics in a big city at night.

The beautiful image don’t seems to draw the doctor’s attention at all. his eyes narrowed drastically, reading the image carefully, as if he is searching for something very important.
Then, something grasps his attention. The doctor raises his arm, with both of his hands wearing black rubber gloves, and starts making gestures as if he is manipulating something invisible in front of him. 
More interestingly, the 3D image of the big blue brain floating above the screen starts to rotates, even being cut into half and showing the cross-section of the brain. As the doctor’s gestures flying in air , different graph and information jumps out or folded swiftly simultaneously.   
‘ Scale 10 electro pulse at G3 in motor cortex, Bryan.’ ‘ Regina, more sodium sulphide composite ! ’ The doctor yells at his people without looking away from the image. Apparently, there must be a surgery under operation with remote controlled robot hands and sensors, in a bacteria free operation room somewhere else, and the surgery must be very important and difficult or it wouldn’t require hundreds of people work together.  
A shimmering tiny object is slowly 



‘We will provide this service to the needed people in government, army, and intelligent agency first.  ’ ‘Within 5 years, this technology ’




   

   




‘ Give him one dose GABA and Dopamine, drop his body temperature to 30 degree.’  ordered the man forcefully. The progress of the recovery of 
   
  
Everything around him is shaded in dark, thick gloom.
I hope  it is the lapse my eyes    My eyes were still sore as if I haven't slept for a thousand year, and my spine twitched badly just like someone had pierced a whole box of steel nails into it. The night wasn't passed easily, to be honest, and anyone knows me knew I am not exaggerating at the strictest sense. I still managed to maintain my marine quality . Time was harder then, but I was, and always will be the kind of man who wouldn’t make the slightest frown when facing the strongest difficulties and most relentless menace. The pain really killed me that I have to grasp the sheet tight that I could reach my feet down to the floor.

I took the pills that I used abusively these days from the fridge to sooth my backache, and I dropped some eye-drops and ointment into my swollen eyes.  I checked my face in the mirror, the eyes were yellowish grey, the pupils blurred and lifeless.  wrinkles and stains scattering across my face and dark veins crawling underneath my transparent skin exposed the profoundest lethargy of a 86 years old man. I had once been young and strong like a howling river, but the river was bleak like desert. life had once been beautiful and precious, but now it mean nothing more than torture.           







Consciousness in terms of physicalism. Science of subjective experience

What is consciousness ? I mean, what is the movie playing in your head everyday since you were aware of it after you were born? The movie not only deliver you the most real visual, audial excitation which you might have experienced when watching a 3D movie, but also gives you all physical sensation that feels like you are in a virtual reality simulator. What is the movie, aside from its multiple sensory stimulations and never-ending story, in which you experienced emotions, desires, and spirituality, and the only protagonist in the story is ‘you’ ? 

Consciousness in the description above can be concisely summarized as ‘the collection of all subjective feelings induced by our experience.’  One may see it absurd and impractical to try to deal with subjective feelings in terms of physicalism because for hundreds of years the two main streams of etymology are idealism and physicalism, which are thought to be incompatible to each other. Despite the success of science, or its under-lied belief in physicalism, it has not provide yet a convincing explanation for the mechanism of our subjective experience. The term subjective explains itself: what happens in mind. How can we measure something that happened in our mind objectively after all?
   
However, the enormous success of scientific methods and its flourishing achievements make physicalism a universal belief. The most faithful believers of physicalism- scientist - have virtually become prophets of modern era.  For any devotee to physicalism, who believes mechanical universe and dismiss super-nature phenomenon, what can be harder to accept than a truth that our minds are not part of this universe ?  

Indeed, neuroscientists have discovered many valuable facts about our brains- how areas of brains correlate to difference mental processes- but they simply have no clue which area is in charge of our subjective feelings. Some scientists thus suggest our mental processes are not only a result from the teamwork of each functional brain areas, but also a quintessential ‘emergent phenomenon’ plays behind the scene. 

Emergent phenomenon is basically a belief that cortical neural network is neither centralized nor localized as brain scientists observed on EEG or fMRI. Cortical neural network works as a decentralized, integrated whole where every ensemble of interactions of nodes and connections represents a specific mental state.    

 Then, how come this two seemingly contradictory views, either of which is supported by considerable evidences, can be fused into a unifying, self-rounded theory of brain ? Some scientists lay their hope in the complex system. Complex systems, which is neither centralized nor decentralized, exhibit hierarchical structures. The hierarchy is not built block by block on account of genetic blue map, but self-organized. The neural network is organized via its agents performing certain simple rules, instead of being dictated or supervised by an external higher being. The most famous rule of such is Hebbien Rule, usually described as ‘fire together, wire together.’

When I was organizing my thoughts about ‘what is consciousness ?’  It was clear to me that most of my inspirations should be credited to Jeff Hawkins and his philosophical but realistic booklet on Intelligence.  His book delineate why intelligence is making prediction and how the cortical neuron might have achieved that. He proposed a group of simultaneously fired neuron may stand for one basic element of our world, and an ensemble of similar elements will be categorized into a superior concept, which will be represented by another group of neuron in a higher layer of neocortex.  For example, our ears first receive a vibration of waves when we listen to a speech. The vibration is then transformed to an electric signal of sound, which to bottom layer of cortex may sounds like alphabet. When it adaptation to the certain combination of alphabets, which always occur consecutively,  it recognizes the combination, the word, with a group of sync-fired neuron. Likewise, it recognizes patterns of combinations of words in to sentence in next layer. Finally, the meaning of the speech is captured through this hierarchical information processing system. Most tasks are well-learned patterns, are trivially predicted in lower layers of cortex, so it never cross our mind. Namely, only fresh, novel stimulation will finally enter the highest layer of neocortex, where our awareness functioned.  Hawkins’s theory is fascinating because it consists with our experiences and easy to understand. It can be easily tested by simply recall you did not really think about the pronunciation of words when you are reading this article. Sometimes you did not even noticed misspellings or grammar error because you subconsciously predicted what the word means. However, we are aware of the spelling of words, and have difficulties understand sentences some times when we are listening to unfamiliar foreign languages. Allow me call the theory above ‘Intelligence theory.’
I think Jeff Hawkins has explained how intelligence works pretty clearly, given how ignorant we are to it previously. He also mentioned he has not figured everything out about consciousness, since it absolutely has something to do with cerebellum, thalamus, and hippocampus, etc. The picture of consciousness is incomplete without a complete understanding of emotions, animal instincts, and their interactions with intelligence. Jeff Hawkins just explained the new brain (neocortex ) , leaving the old brain undiscussed.

To understand subjective experience

Tuesday, December 27, 2016

Neuromodulation


Twenty-five Lessons from Computational Neuromodulation

Peter Dayan

Neural processing of brain is constrained by three problems. A) Computation is is radically distributed, yet amount of point-to-point connections are limited. (B) Bulk of these connections are structurally uniform, lacking target specific physical differences, (C) The brain's structure is relatively fixed, unable to explain the multifarious modalities under different conditions of neural ensemble.
Neuromulators address these problems by their multifarious and broad distribution, By enjoying and specialized receptor types in partially specific enough hey anatomical arrangements, and by their ability to mode activity and sensitivity of neurons and the strength of the plasticity of their synapses. 
As we should see later, architectural and neuromodulatory specifications are frequently integrated. We address Pavlovian (habitual) and model-based learning by reviewing acetycholine (Ach), dopamine (DA), norepinephrine (NE), and serotonin (5-HT). Phasic (transient) dopamine signal resulted from predictive cues provides a formal underpinning for the theory of incentive salience, which is concerned with motivational influences over the attention garnered by such stimuli. We also have noted substructure in the dopamine systems such as the mesocortical dopamine neurons that are excited rather than inhibited by punishment.

Goal-directed control, working memroy
Pavlovian mechanisms that lead to approach or withdrawal to external appetitive and aversive outcomes an predictors might inluence the way theenumeration works. States associated with reward could be more likely to be enomerated under the influence of dopamine and serotonin. Control over working memory can bave both instrumental and Pavlovian components. Basal ganglia could acquire policies that control the gating of information nito working memory using reinforcement learning. 

Uncertainty
The representation, updating and use of uncertainty have become major foci of computational treatments of neural information processing.

Learning
Once case of the link between ignorance and learning arises in the context of auto-associative memory models of the hippocampus. If an input is assessed to be novel, then it is stored; if the input is familiar, the recall process should remove noise from it and/or recall relevant context or associated information. The ability of neuromodulators to control the course of activity by regulating  pathways determines the activity of neurons is a common scheme.  ACh helps regulate oscillations that simultaneously affect multiple sub-regions of the hippocampal formation. It has been suggested that different pathways between theses regions are dominant at different phases of theta.

Model-based predictions and plans are dependent on learning about the structure of the environment and circumstances and outcome contingencies. This should be regulated by predictive uncertainty.
There is structure in a loop connecting cholinergic neuclei to sensory processing and prefrontal cortices.

There's evidence that tonic activity or levels of norepinephrine do indeed increase with predictable reversals in a simple reaction time task, and boosting no norepinephrine (NE) can speed the course of reversal learning. Reversals we shot popular way of inducing change, are normally signaled when actions and choices that used to be rewarded become unproductive or less productive; and actions that were formerly punished or nugatory become worthwhile. One might expect dopamine and serotonin to be involved directly in assessment and realization of reversals. Rapid change is normally the feature of a model based or goal-directed system. With NE, the projection of serotonin and dopamine to the striatum and prefrontal regions having implicated in form of behavioral flexibility such as reversal learning and set shifting.  Mechanism by which bonuses arise in model-based calculations is unknown. 

NE has long been linked with anxiety. NE helps to organize response in stress notably in conjunction with cortisol, a steroid hormone that acts as another neuromodulator. This involves everything from changing storage and usage via glucocorticoids to change the actual style of information processing.

Inference
Selective attention the impact of uncertainty should be governed by the utility associates with what can be discovered; one might expect that much of the inferential certainteed should be highly specific to circumstances of the task. There's evidence for involvement of both ACh and NE in controlling critical aspects of inference at both timescales mentioned above. Cholinergic neuromodulation ass also been implicated in aiding signal detection.
 For normal animals, ACh was substantially released over a short timescale on trials on which animals successfully detected the cue. At fine, sub-second timescale presentation of the cues the phasic release of ACh is related to expectation of the changing circumstances associated with the upcoming reward.

Discussion

For the first, we have seen common motifs such as heterogeneity in space (i.e. different receptor types with different affinities some localize on different systems) and heterogeneity in time ( with phasic and various timescales of tonic release) there's number forms of controls including self-regulation by autoreceptors, complex forms of inter-neuromodulator interaction. Also, loops between prefrontal areas and neuromodulator nuclei which exert mutual influence uptown each other. These, and indeed other functions of the neuromodulators, maybe complicated by corelease of other neurotransmitters and other neuromodulators through the same axons. 

Given the focus on decision-making, the key neuromodulators were dopamine, serotonin, acetylcholine, and norepinephrine, which you presented information about reward punishments and expected and unexpected uncertainty.

Concerning neural processing, the two most important systematic effects are controlling plasticity, and controlling whole pathways. Acetylcholine's influence on thalamocortical versus intracortical interactions. They are so dynamic effects, searches change the gain of competitive, decision-making circuits.

For future, measuring their local concentrations at target zones and selectively manipulate their activity or that of particular receptor types. 

BOLD signal my not would be very effective in helping understanding neuromodulation because the neuromodulators might be able to affect local blood flow directly themselves. There is much work to do to  understand overall networks and systems affects of changes that we know different neuromodulators lead to individual elements in those circuits. This may help understand the aspect of various sorts of heterogeneity. - e.g. What is achieved by the subtle differences within families of receptors and also the rich intertwining of neuromodulators. Model free and model based systems, with the former substantially based on neuromodulators such as dopamine and serotonin, whereas the latter depends on cortical processing.

My thoughts

The effects of neuromodulation is almost everywhere on every function of brains. They are the main mediator of signal processing across multiple spatial and temporal scales. In the review, the paper addressed four principal neuromodulators that implicated on model free, Pavlovian control, model-based, goal-directed control, uncertainty, inference, working memory, and learning. From signal transmission at synapses, to looping between striatum and cortex, their intertwining corelease and heterogeneous targets and receptors allows complex functions of brain be possible with such a limited, and fixed dendritic structures. This paper explained how the three limitations mentioned in the abstract are solved by neuromodulations.  

Sunday, December 25, 2016

What do neuroscientists think of computational neural networks and deep learning?

As a student of neuroscience, I have never heard of computational neural networks. I guess you mean artificial neural networks. I can just speak for myself.
Many theorists believe there is a link between ANN, DL and brains, but to most neuroscientists only believe what evidence tell them. There is not much proof showing ANN and deep learning are how brains work.
The connection between ANNs and brains are nearly non-exsitent. The fact is that ANNs and DL only looks like real neural network from a layman’s perspective. There’s nothing they are alike when you scrutinize the details. For example, the ‘cell’ in ANNs stores values. However, in real cells there are just electrical potentials and the information are full of noises. Besides, the connections in our brains are quite random, and full of recurrent loops in different levels, while in DL the network structures are feed-forward and fixed. In another words, it seems the more successful the ANNs, the less it resembles real neural networks. Even many vision scientists use probabilistic models for visual cortex, it is not explained how a global cost-function minimization, back propagation are achieved neurologically. For deep learning such as deep belief networks, the training is not on-line but off-line (batch training, 2 phase training), which is also very different from our brains.
On the other hand, neuroscientists have fairly clear idea of how grid cells and place cells in hippocampus work to encode spacial memory based on experiments, and the machenism is very different from ANNs. Other successful models for cerebellum are very different from ANNs, either. If ANNs and DL are generic models of brains, then the mechanisms should not be so different.

Friday, March 11, 2016

Emergence of Counsicousness and Attention.


There are only two kinds of things that we need to pay attention. First kind of thing is the thing you want to do, but you can't do. The second kind of thing is the thing you can do, but you don’t want to do. Apart from them, things just pass by like fog and smoke and we don’t see exactly how and when they were done. They are things that we do everyday unconsciously. They are routines and trifles that we’ve done thousands of times that some sort of muscle autonomy just takes over, free us from the tedious labor. Eating while watching TV, talking on a cellphone while driving, Reading papers in the toilet, or listening to music while doing homework. We seem to avoid paying any attention to those tedious and trivial jobs as much as we can even when we nothing better to do. These tasks are duties which we have no choice but doing , otherwise our bodies will stop operating. As long as it agrees with our nature, and nothing goes against it, then our bodies just know how and when to do it right itself. Put it philosophically, our nature conforms nature itself. It follows the laws of thermal dynamics, and try to maximize entropy if we don’t spend energy to stop it.


Attention is inhibition in cognitive sense. We have to inhibit our desires when we can’t do it. We can’t spend freely, wake up late, or smoke as much as we want , even though it’s what life is about. We must inhibit the desires when our knowledge tells us they are detrimental. Therefore, we are painfully paying attention not letting our nature unleashed. On the other hand, we are forced to do things we hate and never get used to on daily bases. We can choose the easy way, but we don’t. because our knowledge tell us they are the necessary vice, otherwise something worse may happen. We have to inhibit our disinclination every second to fight our natural aversion for the longer good.
Inhibition deforms our mental structure, builds up mental tension, stress, and strain, and transforms entropy into potential energy in the mental deformation. Attention is inhibition in phenomenological sense, it is heat and light emitted when potential energy stored in deformation exceeds the energy threshold of the structure. The deformation becomes fracture, and permanently changes the mental default structure. Mental energy discharge is attention, attention leads to mental suffer, mental suffer leads to consciousness, consciousness leads to selfness. Selfness is the crystallization of entropy minimization, the maximization of inhibition, and the singularity of universe.
Inhibition is the essential element of nervous system. It is impossible for nervous system to generate meaningful patterns without incorporating inhibitory signal. Without inhibitory neurons, excitation neurons will fire unbridled, and we will be led to delirious madness and eventual self-destruction. Inhibitory neurons are the miniature self. They defines the character of nervous systems. They are the source of animation, and the starting point of the betrayal of God.


Training Deep NN and Theory of Sleep




It is just an idea, and probably somebody has already been working on it. I still feel the urge to share with you about how I think the Deep NN has inspired me on the evolutional purpose of dreaming, and how theory of dreaming can help us better train Deep NN. However, I must disclaim here, that the following discourse are purely speculative.
NN has been known as a kind of unsupervised learning. i.e., it learns the regularity of data without the help of human expert to label them. The NN then can be trained via ‘backprapagation’, during which the weights of the links, and the value (activity) of the nodes and be updated via optimization of ‘objective function’. Typically, cross entropy is used as objective function. The typical problem one usually confront when training the network is the gradient descent on the landscape of the objective function may fall into local minimum. Several techniques are widely used to help escape from suboptimum solution, such as annealing, adding momentum term, or adding stochastic factor into the training. Here I propose an idea that is inspired by damped oscillation pattern of sleep. The damped oscillation of sleeping stage consists four stages, each of which lasts from 10 to 20 minutes with slight variance depending on individual and various physiological factors. The oscillation is illustrated in the following graph.



Each stage are reported to be responsible for different purpose. The REM -> Stage 4 section can be think for as backward propagation, while the Stage 4 -> REM section represents the forward propagation. The shallow-deep sleep cycle help strengthen the memory and also help internalize the vast amount of transient episodes into deep, invariant knowledges. I will elaborate in the next paragraph. The picture below shows the famous LeNet proposed by Yann LeCun. The net resembles brain structure not only in terms of topology but how the information is processed. Each layer ‘pools’(samples) small patch of the input and forms a feature space that is going to be sampled by a super-layer.You can see the trained middle layers as filters of information. Finally, the every single nodes in the end-layer can distinguish objects exclusively. In broader sense, the deep NN can condense knowledge from a huge batch of unsorted data, be it image, sounds, or movements. The success of Deep NN sheds light on the long coveted universal theory that explains how might brain learns, predict, interpret, and create things. But still, we are very far from finding such a theory, for the anatomical counterparts and physiological mechanism that justifies the NN are not yet approved. Even more, I am neither saying that Deep NN can explains EVERYTHING of brain mechanism nor the universal theory exists.



In general it is agreed upon that sleep is essential for long term memory consolidation, and can enhance cognitive processing. The REM sleep, where we usually dream and can easily be awaken by external disturbance, seems to be playing a critical role of learning. It is instructive that during dreaming our mental activity is actually quite similar to the waking state, and the major difference is the mental process is isolated from body movement. By analyzing the texture of dream, we will find out , unsuprisingly in retrospect, that the function and form conforms each other. The structure of dreams reflects exactly how we stores invariances in the deep, hierarchical structure of filters of different functional level.
In the first cycle of sleep, we dive into deep sleep (stage 1 to stage 4) from aware state. This process correspond to the back propagating phase of LeNet (Input layer to end layer), where in each stage of sleep corresponds to optimization of weights of each layer of NN. It doesn’t mean that the physical counterpart of Deep NN must have four layers, because the definition of stage is artificial and only serve certain purpose. There are not real distinct boundaries between them, it is more like a gradual process. During this half-cycle, the optimization of weights cannot be too good, since it can easily fall into local minimum.
At the deep sleep, end-layer that interfaced with hormone signals produces by glands[1], and other subcortical bodies [2] received instruction and starts to send signal back to sub-layers. These end-layers represents emotional components, and inverse the previous process, starting to predict the possible outcome in the sub feature space given such emotions.Since the inverse problem does not exist a unique solution, we won’t see a playback of our input, but a physically and causally plausible theatre, in which a unpredictable, bizarre, but animated drama is on the show. When we see such a show during sleep, our other parts of brain are aroused, resulting awake-like brain state. The vivid experience are overall endogenous, but its constituents can be novel to each other perceptrons, as the hierarchical brain structure always branch out when goes downward, and the genesis of artifacts is nonlinear due to convoluted operation or quasi-randomized internal driving sources.



During the first REM stage we are exposed to fabricated scenarios that appears arbitrary and illogical, but are strongly emotional and animated. These scenarios are oftentimes consistent in sense of causality and physics, but their motifs and development are totally unpredictable. Take the visual recognition for example. We can generate a face by synthesize features according to the structure of NN, but no single line on of the synthesized image will have the same shape, same shade, and orientation. In a sense the dream provides artificial data let allows us to train the Deep NN again. The subsequent descent of the sleep stage allows optimization for the second time. This helps the NN escape from local optimum of objective function.
This cycle of forward modeling -backpropagation is repeated a few times but not to many for 1) the data novelty deprecate after each cycle. 2) over-fitting. 3) time to wake up. Due to the pyramid shaped hierarchical structure, deeper layers has progressively less perceptrons/ filters, and therefore more prone to over-fitting. This accounts for why the depth of each sleep cycle decrease in light of learning and memory. The lower the hierarchy, the more perceptrons to be trains, and therefore the shallower stages (REM, stage 1, stage 2) occupy the primary proportion in the later part of the whole sleep.
What is the benefit of interfacing the emotional layer with emotional related hormone sources ? It is postulated that the hormones are supervising signal that instruct our brain to learned scenarios that embeds negative emotions ( stress, anxiety, fear, and anger) so that we can handle social relations better. Learning to endure and even utilize our negative emotions during sleep prepares us better to handle frustration and unpredictable difficulties in real world, where most of time we live under certain pressure of survival. It equips us with a inherit sense of crisis, which helped our ancestors survived disasters that those with different mindsets did not.

Footnote:
[1]such as pineal gland, Pituitary gland
[2] like amygdala (emotional memory), mammillary body (recollective, or episodic memory), and hippocampus (spacial memory, STM, LTM).
[3] Other theories of sleep : Activation synthetic theory ; Continual -activation theory ; Reverse learning ; Dreams as excitations of long term memory;