Place your take-home messages here

Download 3.37 Mb.
Size3.37 Mb.
1   ...   49   50   51   52   53   54   55   56   57


Social Expectations

24/7 real life social media launches

expectation to have dialogue

operational- a nightmare

If one customer is discontent he can spread that easily through Internet

Real time

Changing their minds
What is the impact? How to operate differently?

Technology is not good for techology's own sake

what are the things they need - the customers?

Solve the problems.

Call waiting - invented in the 1950s but post phones

Why would I hang up on the person I'm talking to?

Why would u?

Consumer need

Team projects, real consumer business insights,

accelerating technologies.

high growth through VC not always best

companies growing at their own pace

Companies acquire companies

not viable long term strategy to grow too fast
Q - AI- how will that influence business?
These responses to technologies that we discussed above came 10 years ago.
The change will have huge impacts. We need to understand what has happened in the recent history.

Very painful changes. Nobody wants to do it. Must everyone has to.
Important question - How to create a company flexible enough. Consumers and technology changes push business to change.

2:30-2:55: Matthew Kern: Green Semiconductor Formulations
What is green?

- Consider everything

- Have a plan

- Green means \"go\": go out and find a solution
Improve the cleaning process in line and via formation.
EPA: 12 principles of green. John Warner of Warner Babcock

[get EPA list / /principles.html]
Current: Hydroxyl Amide -- explosive, expensive, not green

Goal: same cleaning results, safer, cheaper, low impact.
4 month project, found a result that was cheaper, less explosive, less viscocity, and didn't need wate treatment before disposal, and not on CA's proposition 65 (toxics) list.
Problem: when you change a process in a semiconductor fab there can be months' worth of tests, and massive amount of retrofits.

Group Facilitation: Crash Course

Tony Lyu / Korea
More efficient meetings. Comes from the public sector, specifically UN meetings, that all voices are heard, not just one dominant voice.
Example: 4th of July-- do brainstorming...[people throw out ideas]

* bbq, fireworks, sing the anthem, watch the world cup, beach, parade (of robots!) etc
Next: time (x axis) and originality (y axis)-- as time goes on, moves from easy or obvious answers to the creative and less expected.
Then: group the activities, using various perspectives. Ask questions that spark creativity.
Example: trying to find a solution - \"in South Korea, how many presidents since 1945?\"

brainstorming how to figure this out.

1. Divergence [average length in various countries, any SK politics]

2. Chaos

3. Convergence

OR Premature Closure (latter not a good result. Not a real consensus)
9 micro-skills to use:

Divergence Phase (get people to think)

1. Gather ideas. Don't judge ideas as they come up, distracts from bringing more ideas up. suspend judgement, get quatity.

2. Mirroring- repeat what the other guy said. Don't paraphrase (makes it seem like you're manipulating their idea, appears you're not listening)

3. encourage. People need warming up, especially to help the silent speak up. Best idea- look at them.
Chaos Phase - get people to understand, to explain, to elaborate so that everyone understands

[question - how to know when they switch? when ideas start to overlap, also in part when ideas are being argued]

4. Paraphrase - counterpart to mirroring. \"Am I right in understanding you meant a, b, and c?\"

note: always check back that the original person does, in fact, agree

5. Balance - if one side talks but not the other, it doesn't mean that everyone agrees. Need to hear, ask to hear, from both sides.

6. Make space- if one or a few people are hogging time: choose to ignore some, look at the silent

7. Stacking - give people an order, if three people put up hands, label them1, 2, 3.

8. Tracking - ensure that all discussions on A are complete before moving on to B.
Final Phase: Convergence

9. Intentional silence - time to think, internal dialog. Get people to \"write down what they're thinking\"
8 Degrees of Agreement, from no to yes.

1. Block - worse thing you can do is to Block, to absolutely say \"No\"

2. No - w/ absolve responsibility

3. Formally Disagree, but still agree to go.

4. Standing Aside - \"I don't think this is correct, but I won't agree with you...\"

5. Abstain - neutral

6. Agree - with reservations

7. Minor point of contention

8. Endorsement -
Important Question - to get something done, what's the minimum type of agreement we want? It can depend on cultures, etc. Can start at point 2, but for real consensus at least 3-4.


Alison Lewis: DIY for the non technical person
Background: designer, some CAD, but not electronics. Family - grandmother with crafts in the Smithsonian. Became a web designer. Wanted to bridge the gap between tech, crafts, wanted to have hands-on.
She visited 7 schools, including Parsons School of Design. Markos [Get name], inspiration to go to Parsons.
Picture: Puppies and Kittens
Processes: [and thinking about an audience which is not necessarily using technology, or thinking DIY. A wide audience, age 13 - 60]

1. Focus on their skills: \"if you can do a sewing pattern, you can understand a diagram\"

Familiar toolds, delicate materials, read patterns, math, solder, construction

2. Lifestyle - be inspired by possibilities - example- pillow talk- pillow to plug cellphone into.

3. Draw it out - sketch, diagram, schematics, code

{Idea - a twitter for drawings}

1. can the design stand alone. Picture - fiber optic costume. The design doesn't require the lights.

2. can the tech be made simple? Find solutions / ideas that aren't overwhelming with tech.

3. what's new?

4. what do I need to teach? for example- teaching how to make LED beads.
Then, go shopping:

* supermarket, drugstore, chain fabric, chain hardware, radioshack: use what people can easily get.

* Avoid the 'scary' places = complex, difficult to use websites. If you *must* then give exact part number and link to the part.

* Stay focused - [example- complex craft store - the electronics store is as imposing if you're not used to it] - hear - use a button as the pcb to hold the LED

* No geek speak in craft stores. Talk to them, show them pictures, don't use techie jargon

* Prototype and Photograph. take lots of photos: every step documented. 25 projects in 5 months

* don't stop until it's right. Simplify until it's realistic and do-able.

* red ink - check each step, have a person who can test it, test it.

* Diagram

* Photography - example - showing project in many locations

* Put it out there - iHeartSwitch, threadbangers, instructables, craftzine, volunteer, workshops, birthday presents, etc. Think: how could you scale to teach many people.

Alison's background: she worked at a company- scarf that detected wifi, etc. They started out building everything themselves, and then instead (now) sent out kits for others to use.
Q: interested in advanced technologies? Yes- fashion is interested in tech: \"Epidermal interfaces.\" Wearable and fashion technology are converging. Example: color changing dress by Cute Fashion.

Q: how do you clean -- Electronics and water don't mix? A: technologies are changing, can remove batteries, flexible, etc. See push in Italy for washable technologies.

Q: adoption? Novelty is an issue.

Q: wearable solar


Modeling Brain Dynamics

Rosa Chan, USC.

Project: started as input output properties of the hippocampus

background - cortical prosthesis

Generalized Laguerre-voltera [get name]
idea: a multi-site electrode array to connect long and short term memory, using a VLSI biomimetic model.
1. Three types of neural prostheses

- sensory, see Second Sight company/spinnoff. Case of helping after retinal damage

- motor, after amputation for example

- Cortical prosthesis - her project. Help patients who've lost brain function - a bridge between one and another parts of the brain. Example: movie:memento- lost hippocampus
Need to understand inputs/outputs

example: rat, delayed non match to sample (DMNS) memory task. Similar to remembering where one parked a car.

[Deadwyler's lab @ wake forest university
Two approaches

* I/O Driven models

Available measurements - FMRI, EEG, MEG; Spike trains.

like trying to hear a concert from the outside, listening through the door.

[diagram Brown et al Nature Neurosci 2004]
* Bottom-up approach


IBM Blue Brain,

IBM SyNAPSE Team, [Henry Markham ? in Switzerland]

UCSD Whole Brain,

EONS (elementary objects of the neural system) model at the synapse level

[diagram: Bouteiller et al J of Integrative Neruoscience 2008]
Model from synapse to neuron - EONS...
But neurons are only 10% of the brain. Glia = 90%

Modeling the other brain - take into account Neruon-glia interaction. This has been neglected.
H. is a multi I/O system.

it is dynamical and nonlinear. NOnstationary nonlin dyn underlie spike train transformations.

Also Stoichastic - different spike outputs for the same continuous firing probability
They built a physiologically plausible model, and a stoichastic model.
Then-- she had it implemented on a system.

recording - multichannel processor - spike sorting - program - simulator- output.
Results: a drug (mk801) inhibits H function: they built a real-time facilitation of hippocampal impaired rat.

* They can predict if it's making particular decisions in a MIMO closed loop performance.



*Role of context,


* hormonal fluctuations [corticosterones/stress, estrogens, ghrelin/hunger]
She now has a whole series of data.

Next - moving from rat to monkey model.


David Hutchinson: Feedback in Hierachical Temporal Memory
A topic he likes, not what he researches (which is graphene, etc)
HTM: an interactive heirarchy of levels all waiting...

Will talk about feedback loops

Example: internal letter mixups in words, where people can understand. This will be his model system - how we read words.
Recommends: Hawkins \"On Intelligence\"

3 facets / characteristics of HTM

1 hierarchical reasoning - there is a hierarchy

2 predictions

3 feedback - beyond external and internal world interacting. between internal levels w/in head

You don't close your eyes when you are driving
Example: \"Psychology\" it has a shape, with spikes, etc.

Your brain does some interesting things. Flash word fast. Measure how quickly it could be identified. Research done in 60's.

now \"psychoigy\" or \"psycgology\" therefore brain is looking at overall shape

fastest: \"psychologr\"

Another more interesting thing, your brain picks it up even sooner by chaning the last letter. Seems to start at both ends and work in. Looks at shape.

Also looks at local features. White space, black space, angles.

start at both ends, read inwards, decipher overall shape, real-time sensitization

global and local levels: how does brain talk to each?

[Slide: Real-time sensitization]

So you glance at a word for the first time.

Both local and global feature recognizers jump in.

Local says could be that word or that word, but 90% sure it is NOT that word.

One can say to the other \"try harder\"

Global recognizer takes two words and looks for ups/downs. Sensitizes to particular feature.

finds \"two ups\" and passes that back down to Localizer. results in time delays in what we can recognize.

For example, change a letter to something similar it will take longer than for something completely different.
we have an ability that chimps don't have. Demo: youtube: \"chimps outperform humans at memory tasks\"

Quick demonstration on YouTube nTgeLEWr614

Chimp playing video game. Learned to count. In order. 1 2 3 4 5 .. Scattered .. when 1 pressed, rest blank out. It can do that.

[km note: chimps can/do recognize numerals as such, that \"14\" is larger than \"11\"]

Human tries the same thing.

[Chimp that has been trained for years and years


[Patterns chimp has not seen before

[It was missing 2 there

Yes, just 5 of the 10 numbers.

Point is - a few points.

OK, we have all these feature recognizers.

This means something to me.

I have to work through all these levels.

Points out we have all these levels.
5 chimps in Japan all doing this. Seems to be something there.

There have been other interesting studies.
So this is how [in last 4 minutes] - neural networks and machine learning.

Next thing: real time sensitization.

Seeing now in neural networks. Nodes that interact with each other with different weightings.

You send inputs in, see what you get out, change weightings.

This is a way to get the computer to \"think\" like a brain.

You change the values, so that part of this network can tell other parts what to do.
Q: What does 0.7 mean? 70% probability.

Think of it as 0/1 or -1/1.

Sigmoid curve.

Multiplicative factor of inputs.

Computers using this architecture can detect shapes.

This is the way computers are going, not just line after line of code.

Q: Back to chimp. They can do it better than we can?

If it is a real effect, why is chimp better.

K: Chimps actually understand numbers very well.

5 M&Ms and 7 M&Ms. They do that.

Chimps will always choose the larger plate.

Will also always choose the larger number.

Deeply related to food.

That tree has 14 fruits. That has 17 fruits.
Q: So chimp has more photographic memory?

I think so.

Why might this be the case?

If you are jumping off trees and swinging through branches ...

while we are symbolic.

Q: Dmitry. Trained for a long time.

Q: I have a good example - chess masters have a photographic memory.

If you show them an impossible board position, they can't recall it.

Played so much, they can reconstruct them.
David R: test 2 things. Human eye has natural length of period - latency.

At 15 fps we don't notice choppiness. Birds are way high.
K: Chimps can watch our movies no problem.

David R: Humans would not notice 1/100th

Birds would notice 1/200th of a second flash.

When we look at animal we look at outside edges. Reading inwards is function of that.

Q: Non-human primates only have an approximate number system.

K: Amory Lovins - energy research group - board of great apes research thing.

Has talked about it.

Great apes can watch, but not understand, our movies.
Q: Neural networks can be trained to get same conclusion. Can you find commonalities of two similar networks.

David: You do see emergence of properties. Otherwise, you could just program in that state. It is the emergence of a property bigger than the nodes.

If you shift equivalent nodes ... you have redundancy. See no differerence.

Q: Recognition of a word? Shift letters. Something behind recognizing length ...

Not just one thing on hierarchy. Number of ups/ downs. Shapes of ups/down.
Q: Also true of Hebrew. Japanese.

I suspect so.

Q: Chinese?

Don't know: perhaps shape of white spaces, etc.



Sam Thorpe

A primer on Parkinsons. (not at the neuronal level)

Three famous people: Michael J. Fox, Sergey Brin, muhammad ali.

* 2nd most common neuro disease

* can only diagnose with autopsy

* other diseases have similar symptoms> Parkinsonian diseases, similar appearance, different caues

* 8% genetic, 3% environment, 89% no idea. Or not- this data is contentious.

* one model of P. came from a group of heroin users- a bad batch caused instant P.

* of the 8%, 13 genes, 8 with no pathology reported (we don't know what gene does, or how it affects. just correlated).

* ST worked on PARK1. They almost know what it does.
What is it?

death of neurons in the Substantia nigra. You'll see specific symptoms. These cells make dopamine. w/out it causes movement problems, etc

* by the time diagnosis is possible, 60% of neurons are dead. (60% is Sam's memory. get number)

* one treatment: L-dopa, precursor. Can't give dopamine in directly / doesn't pass BB barrier. See example: MJFox taking, not taking.

* problem: over time you get used to it. Attenuated.

* Deep Brain stimulation - a treatment, not a cure.
Detection: Why do those genes cause P.? Even if we know what the genes do, we don't know why they cause P. i.e. what if it's mitochondria? (it isn't)
Future work

* genome sequencing

* stem cells - not yet, they've tried, failed.

* Sergey Brins data mining approach- see Wired: a way of attacking the problem.

* Genetic manipulation - replace genes, rather than a lifelong treatment.

CD: efforts like folding@home, where they claim they can help things like Alz, etc, can it help with P?

A: certainly can help, what proteins do, etc. Also- can do less sophisticated methods like simply looking at the protein. Need everything on the path: what they look like, how they interact with everything.

Q: Animal models? Computational models?

A: we know so little we work in tiny animals. i.e. Nemotodes, he works with yeast. Very, very difficult: we're at the very basic level for animal models. Computational models? need more data, still very early.

Q could we used FMRI to detect death in the brain?

A: yes, could look, theoretically, but not feasible to have everyone do it

Q: prosthetics? A: see also Rosa Chan.


Share with your friends:
1   ...   49   50   51   52   53   54   55   56   57

The database is protected by copyright © 2020
send message

    Main page