Recent Posts

Pages: 1 2 [3] 4 5 ... 10
21
Home Made Robots / Re: absolutely nothing motor controller
« Last post by ranch vermin on August 19, 2018, 11:49:42 am »
still... havent got it!!  im getting close tho.   so i have to add something more -  the side which is the target, has to be moving with the resistance, and the side which stands for the current position - should be going away from the resistance.

https://www.youtube.com/watch?v=ktVtwCKZnwU&feature=youtu.be


still no luck  - im getting close tho.
22
General Chat / Re: Friday Funny
« Last post by LOCKSUIT on August 17, 2018, 04:10:47 pm »
They say programmers that work on AI are lazy.

Well I'm so lazy, that I don't even program!

I'm the laziest of them all!
23
General Chat / Re: Ten Commandments of Logic
« Last post by RoyMac on August 17, 2018, 12:01:02 pm »
I confess. I'm a big offender at times. lol
24
Robotics News / More efficient security for cloud-based machine learning
« Last post by Tyler on August 17, 2018, 12:00:16 pm »
More efficient security for cloud-based machine learning
17 August 2018, 5:00 am

A novel encryption method devised by MIT researchers secures data used in online neural networks, without dramatically slowing their runtimes. This approach holds promise for using cloud-based neural networks for medical-image analysis and other applications that use sensitive data.

Outsourcing machine learning is a rising trend in industry. Major tech firms have launched cloud platforms that conduct computation-heavy tasks, such as, say, running data through a convolutional neural network (CNN) for image classification. Resource-strapped small businesses and other users can upload data to those services for a fee and get back results in several hours.

But what if there are leaks of private data? In recent years, researchers have explored various secure-computation techniques to protect such sensitive data. But those methods have performance drawbacks that make neural network evaluation (testing and validating) sluggish — sometimes as much as million times slower — limiting their wider adoption.

In a paper presented at this week’s USENIX Security Conference, MIT researchers describe a system that blends two conventional techniques — homomorphic encryption and garbled circuits — in a way that helps the networks run orders of magnitude faster than they do with conventional approaches.

The researchers tested the system, called GAZELLE, on two-party image-classification tasks. A user sends encrypted image data to an online server evaluating a CNN running on GAZELLE. After this, both parties share encrypted information back and forth in order to classify the user’s image. Throughout the process, the system ensures that the server never learns any uploaded data, while the user never learns anything about the network parameters. Compared to traditional systems, however, GAZELLE ran 20 to 30 times faster than state-of-the-art models, while reducing the required network bandwidth by an order of magnitude.

One promising application for the system is training CNNs to diagnose diseases. Hospitals could, for instance, train a CNN to learn characteristics of certain medical conditions from magnetic resonance images (MRI) and identify those characteristics in uploaded MRIs. The hospital could make the model available in the cloud for other hospitals. But the model is trained on, and further relies on, private patient data. Because there are no efficient encryption models, this application isn’t quite ready for prime time.

“In this work, we show how to efficiently do this kind of secure two-party communication by combining these two techniques in a clever way,” says first author Chiraag Juvekar, a PhD student in the Department of Electrical Engineering and Computer Science (EECS). “The next step is to take real medical data and show that, even when we scale it for applications real users care about, it still provides acceptable performance.”

Co-authors on the paper are Vinod Vaikuntanathan, an associate professor in EECS and a member of the Computer Science and Artificial Intelligence Laboratory, and Anantha Chandrakasan, dean of the School of Engineering and the Vannevar Bush Professor of Electrical Engineering and Computer Science.

Maximizing performance

CNNs process image data through multiple linear and nonlinear layers of computation. Linear layers do the complex math, called linear algebra, and assign some values to the data. At a certain threshold, the data is outputted to nonlinear layers that do some simpler computation, make decisions (such as identifying image features), and send the data to the next linear layer. The end result is an image with an assigned class, such as vehicle, animal, person, or anatomical feature.

Recent approaches to securing CNNs have involved applying homomorphic encryption or garbled circuits to process data throughout an entire network. These techniques are effective at securing data. “On paper, this looks like it solves the problem,” Juvekar says. But they render complex neural networks inefficient, “so you wouldn’t use them for any real-world application.”

Homomorphic encryption, used in cloud computing, receives and executes computation all in encrypted data, called ciphertext, and generates an encrypted result that can then be decrypted by a user. When applied to neural networks, this technique is particularly fast and efficient at computing linear algebra. However, it must introduce a little noise into the data at each layer. Over multiple layers, noise accumulates, and the computation needed to filter that noise grows increasingly complex, slowing computation speeds.

Garbled circuits are a form of secure two-party computation. The technique takes an input from both parties, does some computation, and sends two separate inputs to each party. In that way, the parties send data to one another, but they never see the other party’s data, only the relevant output on their side. The bandwidth needed to communicate data between parties, however, scales with computation complexity, not with the size of the input. In an online neural network, this technique works well in the nonlinear layers, where computation is minimal, but the bandwidth becomes unwieldy in math-heavy linear layers.

The MIT researchers, instead, combined the two techniques in a way that gets around their inefficiencies.

In their system, a user will upload ciphertext to a cloud-based CNN. The user must have garbled circuits technique running on their own computer. The CNN does all the computation in the linear layer, then sends the data to the nonlinear layer. At that point, the CNN and user share the data. The user does some computation on garbled circuits, and sends the data back to the CNN. By splitting and sharing the workload, the system restricts the homomorphic encryption to doing complex math one layer at a time, so data doesn’t become too noisy. It also limits the communication of the garbled circuits to just the nonlinear layers, where it performs optimally.

“We’re only using the techniques for where they’re most efficient,” Juvekar says.

Secret sharing

The final step was ensuring both homomorphic and garbled circuit layers maintained a common randomization scheme, called “secret sharing.” In this scheme, data is divided into separate parts that are given to separate parties. All parties synch their parts to reconstruct the full data.

In GAZELLE, when a user sends encrypted data to the cloud-based service, it’s split between both parties. Added to each share is a secret key (random numbers) that only the owning party knows. Throughout computation, each party will always have some portion of the data, plus random numbers, so it appears fully random. At the end of computation, the two parties synch their data. Only then does the user ask the cloud-based service for its secret key. The user can then subtract the secret key from all the data to get the result.

“At the end of the computation, we want the first party to get the classification results and the second party to get absolutely nothing,” Juvekar says. Additionally, “the first party learns nothing about the parameters of the model.”

Source: MIT News - CSAIL - Robotics - Computer Science and Artificial Intelligence Laboratory (CSAIL) - Robots - Artificial intelligence

Reprinted with permission of MIT News : MIT News homepage



Use the link at the top of the story to get to the original article.
25
XKCD Comic / XKCD Comic : Equations
« Last post by Tyler on August 17, 2018, 12:00:14 pm »
Equations
17 August 2018, 5:00 am

All electromagnetic equations: The same as all fluid dynamics equations, but with the 8 and 23 replaced with the permittivity and permeability of free space, respectively.

Source: xkcd.com

26
Robotics News / Design tool reveals a product’s many possible performance tradeoffs
« Last post by Tyler on August 16, 2018, 12:01:06 pm »
Design tool reveals a product’s many possible performance tradeoffs
15 August 2018, 3:00 pm

MIT researchers have developed a tool that makes it much easier and more efficient to explore the many compromises that come with designing new products.

Designing any product — from complex car parts down to workaday objects such as wrenches and lamp stands — is a balancing act with conflicting performance tradeoffs. Making something lightweight, for instance, may compromise its durability.

To navigate these tradeoffs, engineers use computer-aided design (CAD) programs to iteratively modify design parameters — say, height, length, and radius of a product — and simulate the results for performance objectives to meet specific needs, such as weight, balance, and durability.

But these programs require users to modify designs and simulate the results for only one performance objective at a time. As products usually must meet multiple, conflicting performance objectives, this process becomes very time-consuming.

In a paper presented at this week’s SIGGRAPH conference, researchers from the Computer Science and Artificial Intelligence Laboratory (CSAIL) describe a visualization tool for CAD that, for the first time, lets users instead interactively explore all designs that best fit multiple, often-conflicting performance tradeoffs, in real time.

The tool first calculates optimal designs for three performance objectives in a precomputation step. It then maps all those designs as color-coded patches on a triangular graph. Users can move a cursor in and around the patches to prioritize one performance objective or another. As the cursor moves, 3-D designs appear that are optimized for that exact spot on the graph.

“Now you can explore the landscape of multiple performance compromises efficiently and interactively, which is something that didn’t exist before,” says Adriana Schulz, a CSAIL postdoc and first author on the paper.

Co-authors on the paper are Harrison Wang, a graduate student in mechanical engineering; Eitan Grinspun, an associate professor of computer science at Columbia University; Justin Solomon, an assistant professor in electrical engineering and computer science; and Wojciech Matusik, an associate professor in electrical engineering and computer science.

The new work builds off a tool, InstantCAD, developed last year by Schulz, Matusik, Grinspun, and other researchers. That tool let users interactively modify product designs and get real-time information on performance. The researchers estimated that tool could reduce the time of some steps in designing complex products to seconds or minutes, instead of hours.

However, a user still had to explore all designs to find one that satisfied all performance tradeoffs, which was time-consuming. This new tool represents “an inverse,” Schulz says: “We’re directly editing the performance space and providing real-time feedback on the designs that give you the best performance. A product may have 100 design parameters … but we really only care about how it behaves in the physical world.”

In the new paper, the researchers home in on a critical aspect of performance called the “Pareto front,” a set of designs optimized for all given performance objectives, where any design change that improves one objective worsens another objective. This front is usually represented in CAD and other software as a point cloud (dozens or hundreds of dots in a multidimensional graph), where each point is a separate design. For instance, one point may represent a wrench optimized for greater torque and less mass, while a nearby point will represent a design with slightly less torque, but more mass.

Engineers laboriously modify designs in CAD to find these Pareto-optimized designs, using a fair amount of guesswork. Then they use the front’s visual representation as a guideline to find a product that meets a specific performance, considering the various compromises.

The researchers’ tool, instead, rapidly finds the entire Pareto front and turns it into an interactive map. Inputted into the model is a product with design parameters, and information about how those parameters correspond to specific performance objectives.

The model first quickly uncovers one design on the Pareto front. Then, it uses some approximation calculations to discover tiny variations in that design. After doing that a few times, it captures all designs on the Pareto front. Those designs are mapped as colored patches on a triangular graph, where each patch represents one Pareto-optimal design, surrounded by its slight variations. Each edge of the graph is labeled with a separate performance objective based on the input data.

In their paper, the researchers tested their tool on various products, including a wrench, bike frame component, and brake hub, each with three or four design parameters, as well as a standing lamp with 21 design parameters.

With the lamp, for example, all 21 parameters relate to the thickness of the lamp’s base, height and orientation of its stand, and length and orientation of three elbowed beams attached to the top that hold the light bulbs. The system generated designs and variations corresponding to more than 50 colored patches reflecting a combination of three performance objectives: focal distance, stability, and mass. Placing the cursor on a patch closer to, say, focal distance and stability generates a design with a taller, straighter stand and longer beams oriented for balance. Moving the cursor farther from focal distance and toward mass and stability generates a design with thicker base and a shorter stand and beams, tilted at different angles.

Some designs change quite dramatically around the same region of performance tradeoffs and even within the same cluster. This is important from an engineer’s perspective, Schulz says. “You’re finding two designs that, even though they’re very different, they behave in similar ways,” she says. Engineers can use that information “to find designs that are actually better to meet specific use cases.”

The work was supported by the Defense Advanced Research Projects Agency, the Army Research Office, the Skoltech-MIT Next Generation Program, and the National Science Foundation.

Source: MIT News - CSAIL - Robotics - Computer Science and Artificial Intelligence Laboratory (CSAIL) - Robots - Artificial intelligence

Reprinted with permission of MIT News : MIT News homepage



Use the link at the top of the story to get to the original article.
27
XKCD Comic / XKCD Comic : Repair or Replace
« Last post by Tyler on August 16, 2018, 12:01:05 pm »
Repair or Replace
15 August 2018, 5:00 am

Just make sure all your friends and family are out of the car, or that you've made backup friends and family at home.

Source: xkcd.com

28
General AI Discussion / Re: Simple AI Website to let you know what is AI
« Last post by LOCKSUIT on August 16, 2018, 01:23:19 am »
more gifts from the gods:

https://www.autodraw.com/
http://www.cs.toronto.edu/~ilya/rnn.html#
http://www.cs.toronto.edu/~graves/handwriting.html
http://make3d.cs.cornell.edu/






can reconstruct 3d scene video from audio and vice versa

image editing with GANs....morph, stitch, etc

HER holding facts and retrieve em if need

object interpolation cool

recreate video games with no game engine

impersonate anyone with ai made 3d face and shoulder as you talk as them in real time
30
General Project Discussion / Re: The last invention.
« Last post by korrelan on August 15, 2018, 11:06:53 pm »
Ah I see they call it the motion after-effect, not the persistence of motion… I’d never heard of this before… sweet.

 :)

@Lock

It shows that the visual mechanism sensing motion has a kind of inertia, and motion is not sensed/ calculated on a frame to frame basis.  So the visual cortex neurons representing sensed motion at specific regions in the visual field are firing for longer durations dependent on the length of stimulation… probably not relevant to your AGI.

 :)
Pages: 1 2 [3] 4 5 ... 10

Something crazy is happening to me with my work
by ranch vermin (General Hardware Talk)
Today at 06:17:29 am
XKCD Comic : Dark Matter Candidates
by Tyler (XKCD Comic)
August 20, 2018, 12:00:55 pm
outline from gadient mask
by yotamarker (General AI Discussion)
August 19, 2018, 03:09:03 pm
absolutely nothing motor controller
by ranch vermin (Home Made Robots)
August 19, 2018, 11:49:42 am
Friday Funny
by LOCKSUIT (General Chat)
August 17, 2018, 04:10:47 pm
Ten Commandments of Logic
by RoyMac (General Chat)
August 17, 2018, 12:01:02 pm
XKCD Comic : Equations
by Tyler (XKCD Comic)
August 17, 2018, 12:00:14 pm
XKCD Comic : Repair or Replace
by Tyler (XKCD Comic)
August 16, 2018, 12:01:05 pm

Users Online

64 Guests, 0 Users

Most Online Today: 86. Most Online Ever: 208 (August 27, 2008, 09:36:30 am)

Articles