Earlier this year, SmartThings announced a new program called Certification by Similarity (CbS) within its Works with SmartThings (WWST) partner program. This program is designed for Smart Home OEM’s to certify portfolios of products, rather than certifying each product one by one.
Additionally, we released a new developer feature called Product Cloning, which allows partners to input the details of one product and copy for all similar products.
Many of our device manufacturing partners have a portfolio of devices with the same smart technology on the inside but come in many different shapes, colors, and other variations. We are making it easier and more cost-efficient than ever to certify the entire portfolio.
Upon releasing this new certification option, we have heard excellent reviews from our partners. Read on to discover how to take advantage of this new offering.
“We love SmartThings’ new Developer Center and used the Certification Console to quickly certify our entire portfolio of devices.”
Antonin Carre, Product Manager Partnerships & Integrations for WiZ at Signify
Product Cloning
SmartThings Product Cloning allows you to enter information for one product and clone it to generate multiple entries for all similar products. The similar products include all the critical information — all you need to do is update the unique identifiers, like the Matter Product ID and Model Number.
Easily copy product details and enter whole product lines
Save time and get certified faster
Get Started
How to clone a product:
Visit the SmartThings Certification Console.
Navigate to the Products page (second icon from the top on the left).
Enter your product details for one product and save it.
Navigate back to the Products page. Under the Actions menu on the product, you have an option to create multiple products at once with Product Cloning.
You can add up to 10 clones with the option to enter the unique identifiers. Once created, the clones appear in your Products list; you can open them up and modify any of the details before submitting for Works with SmartThings Certification.
WWST Certification by Similarity
Certification by Similarity (CbS) allows you to reduce your WWST certification time and cost by identifying related products with similar core functionality but with different model numbers and minor differences like colors, shapes, or regional variations.
Example Certification by Similarity use case with a portfolio of RGBW, RGB, and white balance light bulbs:
Start by submitting a primary product for certification that is a superset of all the portfolio features — such as one of the RGBW light bulbs. Once the primary product testing is complete, the similar devices — such as the RGB and white balance light bulbs — can get fast-tracked through certification, for free. Additional devices can be either submitted at the same time as the primary product or at a later date. All you need to do is submit the product information and link it with the primary product. We verify that the features are the same as your primary product and grant you certification.
Devices that Can be Considered for CbS
The following criteria must be met for the devices to be considered for CbS:
Note: Cloud Connected Devices may have different Device Profiles and still be considered for Certification by Similarity. This is possible only if the Capabilities for similar products are a subset of the primary product. If a similar device has additional Capabilities, partial testing is required.
Learn more about CbS in the Developer Documentation and Certification Console.
Get Started
How to submit similar products for WWST:
Visit the SmartThings Certification Console.
Navigate to the Certifications page.
Submit your primary product for certification.
Navigate back to the Certifications page and select the target similar product to certify. You now have the option to link this submission with the primary product.
Select the associated primary product and submit your product for certification.
Certification by Similarity FAQs
How is the primary product determined?
→ The primary product has all the Capabilities of other devices in the group. In this example, Product 4 or Product 5 could be the primary.
How can we guarantee CbS will be approved before submitting for WWST certification?
→ We recommend following the guidelines in the Developer Documentation. The WWST team makes the final decision after reviewing your submission.
Should I submit the primary product and wait for it to be fully certified before submitting secondary devices to be considered for CbS, or can all of the products be submitted together?
→ When submitting products, you do not need to wait. You can submit the primary product and similar products at the same time.
I have multiple brands, including some that have the same hardware and firmware. Can CbS be extended to these multiple brands?
→ In order to be considered for CbS, products must contain the same brand. View our documentation to review the CbS program requirements.
How does the publication / timing work for CbS devices compared to the primary product?
→ See our Publish Your Device guide on publication/timing.
Want to integrate your device(s) with SmartThings? Visit our Developer Center to get started and access Product Cloning and Certification by Similarity tools.
Federal Lawsuit Raises Questions About AI-Generated Content
A federal lawsuit over Minnesota’s “Use of Deep Fake Technology to Influence An Election” law is now directly dealing with the influence of AI. In a recent filing, attorneys challenging the law say an affidavit submitted to support it shows signs of containing AI-generated text.
AI-Generated Text in Affidavit?
Attorney General Keith Ellison asked Stanford Social Media Lab founding director Jeff Hancock to make the submission, but the document filed includes non-existent sources that seem to have been hallucinated by ChatGPT or another large language model (LLM).
Questionable Sources
Hancock’s affidavit cites a 2023 study published in the Journal of Information Technology & Politics titled “The Influence of Deepfake Videos on Political Attitudes and Behavior.” However, according to the Minnesota Reformer, there is no record of that study in the Journal of Information Technology & Politics or any other publication.
Another source cited in Hancock’s declaration, “Deepfakes and the Illusion of Authenticity: Cognitive Processes Behind Misinformation Acceptance,” doesn’t seem to exist either.
Lawyers Raise Concerns
Lawyers for Minnesota state Rep. Mary Franson and Christopher Khols — a conservative YouTuber who goes by Mr Reagan — wrote in a filing, “The citation bears the hallmarks of being an artificial intelligence (AI) ‘hallucination,’ suggesting that at least the citation was generated by a large language model like ChatGPT.”
“Plaintiffs do not know how this hallucination wound up in Hancock’s declaration, but it calls the entire document into question, especially when much of the commentary contains no methodology or analytic logic whatsoever.”
Conclusion
The controversy surrounding the use of AI-generated content in the affidavit raises questions about the reliability of the document and the potential influence of AI on the legal system. As AI technology continues to evolve, it is essential to ensure that it is used responsibly and transparently.
FAQs
Q: What is the Minnesota “Use of Deep Fake Technology to Influence An Election” law?
A: The law is a federal lawsuit aimed at regulating the use of deep fake technology to influence elections.
Q: Who is Jeff Hancock?
A: Jeff Hancock is the founding director of the Stanford Social Media Lab.
Q: What is ChatGPT?
A: ChatGPT is a large language model that can generate human-like text.
Q: What are the implications of AI-generated content in the legal system?
A: The implications of AI-generated content in the legal system are still being explored, but it raises questions about the reliability and transparency of documents and the potential influence of AI on legal proceedings.
This article is part of the Circuits thread, a collection of short articles and commentary by an open scientific collaboration delving into the inner workings of neural networks.
Zoom In: An Introduction to Circuits
Curve Detectors
The first few articles of the Circuits project will be focused on early vision in InceptionV1 — for our purposes, the five convolutional layers leading up to the third pooling layer.
For our purposes, we’ll consider early vision to be the first five layers. Click on a layer to jump to section.
Studying early vision has two major advantages as a starting point in our investigation.
Firstly, it’s particularly easy to study:
it’s close to the input, the circuits are only a few layers deep, there aren’t that many different neurons,It’s common for vision models to have on the order of 64 channels in their initial convolutional layers, which are applied at many spatial positions. So while there are many neurons, the number of unique neurons is orders of magnitude smaller. and the features seem quite simple.
Secondly, early vision seems most likely to be universal: to have the same features and circuits form across different architectures and tasks.
Before we dive into detailed explorations of different parts of early vision, we wanted to give a broader overview of how we presently understand it.
This article sketches out our understanding, as an annotated collection of what we call “neuron groups.”
We also provide illustrations of selected circuits at each layer.
By limiting ourselves to early vision, this article “only” considers the first 1,056 neurons of InceptionV1.We will not discuss the “bottleneck” neurons in mixed3a/mixed3b, which we generally think of as low-rank connections to the previous layer.
But our experience is that a thousand neurons is more than enough to be disorienting when one begins studying a model.
Our hope is that this article will help readers avoid this disorientation by providing some structure and handholds for thinking about them.
Playing Cards with Neurons
Dmitri Mendeleev is often accounted to have discovered the Periodic Table by playing “chemical solitaire,” writing the details of each element on a card and patiently fiddling with different ways of classifying and organizing them.
Some modern historians are skeptical about the cards, but Mendeleev’s story is a compelling demonstration of that there can be a lot of value in simply organizing phenomena, even when you don’t have a theory or firm justification for that organization yet.
Mendeleev is far from unique in this.
For example, in biology, taxonomies of species preceded genetics and the theory of evolution giving them a theoretical foundation.
Our experience is that many neurons in vision models seem to fall into families of similar features.
For example, it’s not unusual to see a dozen neurons detecting the same feature in different orientations or colors.
Perhaps even more strikingly, the same “neuron families” seem to recur across models!
Of course, it’s well known that Gabor filters and color contrast detectors reliably comprise neurons in the first layer of convolutional neural networks, but we were quite surprised to see this generalize to later layers.
This article shares our working categorization of units in the first five layers of InceptionV1 into neuron families.
These families are ad-hoc, human defined collections of features that seem to be similar in some way.
We’ve found these helpful for communicating among ourselves and breaking the problem of understanding InceptionV1 into smaller chunks.
While there are some families we suspect are “real”, many others are categories of convenience, or categories we have low-confidence about.
The main goal of these families is to help researchers orient themselves.
In constructing this categorization, our understanding of individual neurons was developed by looking at feature visualizations, dataset examples, how a feature is built from the previous layer, how it is used by the next layer, and other analysis.
It’s worth noting that the level of attention we’ve given to individual neurons varies greatly: we’ve dedicated entire forthcoming articles to detailed analysis some of these units, while many others have only received a few minutes of cursory investigation.
In some ways, our categorization of units is similar to Net Dissect, which correlates neurons with a pre-defined set of features and groups them into categories like color, texture, and object.
This has the advantage of removing subjectivity and being much more scalable.
At the same time, it also has downsides: correlation can be misleading and the pre-defined taxonomy may miss the true feature types. Net Dissect was very elegant work which advanced our ability to systematically talk about network features.
However, to understand the differences between correlating features with a pre-defined taxonomy and individually studying them, it may be illustrative to consider how it classifies some features.
Net Dissect doesn’t include the canonical InceptionV1, but it does include a variant of it.
Glancing through their version of layer mixed3b we see many units which appear from dataset examples likely to be familiar feature types like curve detectors, divot detectors, boundary detectors, eye detector, and so forth, but are classified as weakly correlated with another feature — often objects that it seems unlikely could be detected at such an early layer.
Or in another fun case, there is a feature (372) which is most correlated with a cat detector, but appears to be detecting left-oriented whiskers!
In particular, if we expect models to have novel, unanticipated features — for example, high-low frequency detectors — the fact that they are unanticipated makes them impossible to include in a set of pre-defined features.
The only way to discover them is the laborious process of manually investigating each feature.
In the future, you could imagine hybrid approaches, where a human investigator is saved time by having many features sorted into a (continually growing) set of known features, especially if the universality hypothesis holds.
Caveats
This is a broad overview and our understanding of many of these units is low-confidence. We fully expect, in retrospect, to realize we misunderstood some units and categories.
Many neuron groups are catch-all categories or convenient organizational categories that we don’t think reflect fundamental structure.
Even for neuron groups we suspect do reflect a fundamental structure (eg. some can be recovered from factorizing the layer’s weight matrices) the boundaries of these groups can be blurry and some neurons inclusion involve judgement calls.
Presentation of Neurons
In order to talk about neurons, we need to somehow represent them.
While we could use neuron indices, it’s very hard to keep hundreds of numbers straight in one’s head.
Instead, we use feature visualizations, optimized images which highly stimulate a neuron.
Our feature visualization is done with the lucid library.
We use small amounts of transformation robustness when visualizing the first few layers, because it has a larger proportional affect on their small receptive fields, and increase as we move to higher layers.
For low layers, we use L2 regularization to push pixels towards gray.
For the first layer, we follow the convention of other papers and just show the weights, which for the special case of the first layer are equivalent to feature visualization with the right L2 penalty.
When we represent a neuron with a feature visualization, we don’t intend to claim that the feature visualization captures the entirety of the neuron’s behavior.
Rather, the role of a feature visualization is like a variable name in understanding a program.
It replaces an arbitrary number with a more meaningful symbol .
Presentation of Circuits
Although this article is focused on giving an overview of the features which exist in early vision, we’re also interested in understanding how they’re computed from earlier features.
To do this, we present circuits consisting of a neuron, the units it has the strongest (L2 norm) weights to in the previous layer, and the weights between them.
Some neurons in mixed3a and mixed3b are in branches consisting of a “bottleneck” 1×1 conv that greatly reduces the number of channels followed by a 5×5 conv. Although there is a ReLU between them, we generally think of them as a low rank factorization of a single weight matrix and visualize the product of the two weights. Additionally, some neurons in these layers are in a branch consisting of maxpooling followed by a 1×1 conv; we present these units as their weights replicated over the region of their maxpooling.
In some cases, we’ve also included a few neurons that have weaker connections if they seem to have particular pedagogical value; in these cases, we’ve mentioned doing so in the caption.
Neurons are visually displayed by their feature visualizations, as discussed above.
Weights are represented using a color map with red as positive and blue as negative.
For example, here is a circuit of a circle detecting unit in mixed3a being assembled from earlier curves and a primitive circle detector.
We’ll discuss this example in more depth later.
Click on the feature visualization of any neuron to see more weights!
At any point, you can click on a neuron’s feature visualization to see its weights to the 50 neurons in the previous layer it is most connected to (that is, how it assembled from the previous layer, and also the 50 neurons in the next layer it is most connected to (that is, how it is used going forward).
This allows further investigation, and gives you an unbiased view of the weights if you’re concerned about cherry-picking.
conv2d0
The first conv layer of every vision model we’ve looked at is mostly comprised of two kinds of features: color-contrast detectors and Gabor filters.
InceptionV1′s conv2d0 is no exception to this rule, and most of its units fall into these categories.
In contrast to other models, however, the features aren’t perfect color contrast detectors and Gabor filters.
For lack of a better word, they’re messy.
We have no way of knowing, but it seems likely this is a result of the gradient not reaching the early layers very well during training.
Note that InceptionV1 predated the adoption of modern techniques like batch norm and Adam, which make it much easier to train deep models well.
If we compare to the TF-Slim rewrite of InceptionV1, which does use BatchNorm, we see crisper features.
The weights for the units in the first layer of the TF-Slim version of InceptionV1, which adds BatchNorm. (Units are sorted by the first principal component of the adjacency matrix between the first and second layers.) These features are typical of a well trained conv net. Note how, unlike the canonical InceptionV1, these units have a crisp division between black and white Gabors, color Gabors, color-contrast units and color center-surround units.
One subtlety that’s worth noting here is that Gabor filters almost always come in pairs of weights which are negative versions of each other, both in InceptionV1 and other vision models.
A single Gabor filter can only detect edges at some offsets, but the negative version fills in holes, allowing for the formation of complex Gabor filters in the next layer.
Show all 28 neurons.
Collapse neurons.
Gabor filters are a simple edge detector, highly sensitive to the alignment of the edge. They’re almost universally found in the fist layer of vision models. Note that Gabor filters almost always come in pairs of negative reciprocals.
Show all 27 neurons.
Collapse neurons.
These units detect a color one side of their receptive field, and the opposite color on the other side. Compare to later color contrast (conv2d1, conv2d2, mixed3a, mixed3b).
Units that don’t fit in another category.
conv2d1
In conv2d1, we begin to see some of the classic complex cell features of visual neuroscience.
These neurons respond to similar patterns to units in conv2d0, but are invariant to some changes in position and orientation.
Complex Gabors:
A nice example of this is the “Complex Gabor” feature family.
Like simple Gabor filters, complex Gabors detect edges.
But unlike simple Gabors, they are relatively invariant to the exact position of the edge or which side is dark or light.
This is achieved by being excited by multiple Gabor filters in similar orientations — and most critically, by being excited by “reciprocal Gabor filters” that detect the same pattern with dark and light switched.
This can be seen as an early example of the “union over cases” motif.
All neurons in the previous layer with at least 30% of the max weight magnitude are shown,
both positive (excitation) and negative (inhibition).
Click on a neuron to see its forwards and backwards weights.
Note that conv2d1 is a 1×1 convolution, so there’s only a single weight — a single line, in this diagram — between each channel in the previous and this one.
There is a pooling layer between them, so the features it connects to are pooled versions of the previous layer rather than original features.
This plays an important role in determining the features we observe: in models with larger convolutions in their second layer, we often see a jump to crude versions of the larger more complex features we’ll see in the following layers.
In addition to Complex Gabors, we see a variety of other features, including
more invariant color contrast detectors, Gabor-like features that are less selective for a single orientation, and lower-frequency features.
Show all 17 neurons.
Collapse neurons.
These units seem to respond to lower-frequency edge patterns, but we haven’t studied them very carefully.
Show all 11 neurons.
Collapse neurons.
These units respond to edges stimuli, but seem to respond to a wider range of orientations, and also respond to color contrasts that align with the edge. We haven’t studied them very carefully.
These units detect a color on one side of the receptive field, and a different color on the opposite side. Composed of lower-level color contrast detectors, they often respond to color transitions in a range of translation and orientation variations. Compare to earlier color contrast (conv2d0) and later color contrast (conv2d2, mixed3a, mixed3b).
These units respond to mixtures of colors without an obvious strong spatial structure preference.
Like Gabor Filters, but fairly invariant to the exact position, formed by adding together multiple Gabor detectors in the same orientation but different phases. We call these ‘Complex’ after complex cells in neuroscience.
Two of these units seem to track brightness (bright vs dark), while the other two units seem to mostly track hue, dividing the space of hues between them. One responds to red/orange/yellow, while the other responds to purple/blue/turqoise. Unfortunately, their circuits seem to heavily rely on the existence of a Local Response Normalization layer after conv2d0, which makes it hard to reason about.
Units that don’t fit in another category.
This unit detects Gabor patterns in two orthogonal directions, selecting for a “hatch” pattern.
conv2d2
In conv2d2 we see the emergence of very simple shape predecessors.
This layer sees the first units that might be described as “line detectors”, preferring a single longer line to a Gabor pattern and accounting for about 25% of units.
We also see tiny curve detectors, corner detectors, divergence detectors, and a single very tiny circle detector.
One fun aspect of these features is that you can see that they are assembled from Gabor detectors in the feature visualizations, with curves being built from small piecewise Gabor segments.
All of these units still moderately fire in response to incomplete versions of their feature, such as a small curve running tangent to the edge detector.
Since conv2d2 is a 3×3 convolution, our understanding of these shape precursor features (and some texture features) maps to particular ways Gabor and lower-frequency edges are being spatially assembled into new features.
At a high-level, we see a few primary patterns:
We also begin to see various kinds of texture and color detectors start to become a major constituent of the layer, including color-contrast and color center surround features, as well as Gabor-like, hatch, low-frequency and high-frequency textures.
A handful of units look for different textures on different sides of their receptive field.
These units detect a color on one side of the receptive field, and a different color on the opposite side. Composed of lower-level color contrast detectors, they often respond to color transitions in a range of translation and orientation variations. Compare to earlier color contrast (conv2d0, conv2d1) and later color contrast (mixed3a, mixed3b).
These units are beginning to look for a single primary line. Some look for different colors on each side. Many exhibit “combing” (small perpendicular lines along the main one), a very common but not presently understood phenomenon in line-like features across vision models. Compare to shifted lines and later lines (mixed3a).
Show all 16 neurons.
Collapse neurons.
These units look for edges “shifted” to the side of the receptive field instead of the middle. This may be linked to the many 1×1 convs in the next layer. Compare to lines (non-shifted) and later lines (mixed3a).
Show all 15 neurons.
Collapse neurons.
A broad category of units detecting repeating local structure.
Show all 14 neurons.
Collapse neurons.
Catch-all category for all other units.
Show all 13 neurons.
Collapse neurons.
Show all 12 neurons.
Collapse neurons.
Show all 12 neurons.
Collapse neurons.
Show all 12 neurons.
Collapse neurons.
Like complex Gabor units from the previous layer, but larger. They’re probably starting to be better described as a texture.
These units look for different textures on opposite sides of their receptive field. One side is typically a Gabor pattern.
These units detect Gabor patterns in two orthogonal directions, selecting for a “hatch” pattern.
Several units look for mixtures of colors but seem indifferent to their organization.
These units detect two Gabor patterns which meet at apprixmately 90 degrees, causing them to respond to corners.
These units detect lines diverging from a point.
mixed3a
mixed3a has a significant increase in the diversity of features we observe.
Some of them — curve detectors and high-low frequency detectors — were discussed in Zoom In
and will be discussed again in later articles in great detail.
But there are some really interesting circuits in mixed3a which we haven’t discussed before,
and we’ll go through a couple selected ones to give a flavor of what happens at this layer.
Black & White Detectors: One interesting property of mixed3a is the emergence of “black and white” detectors, which detect the absence of color.
Prior to mixed3a, color contrast detectors look for transitions of a color to near complementary colors (eg. blue vs yellow).
From this layer on, however, we’ll often see color detectors which compare a color to the absence of color.
Additionally, black and white detectors can allow the detection of greyscale images, which may be correlated with ImageNet categories (see
4a:479 which detects black and white portraits).
The circuit for our black and white detector is quite simple:
almost all of its large weights are negative, detecting the absence of colors.
Roughly, it computes NOT(color_feature_1 OR color_feature_2 OR ...).
The sixteen strongest magnitude weights to the previous layer are shown.
For simplicity, only one spatial weight for positive and negative have been shown, but they all have almost identical structure.
Click on a neuron to see its forwards and backwards weights.
Small Circle Detector: We also see somewhat more complex shapes in mixed3a. Of course, curves (which we discussed in Zoom In) are a prominent example of this.
But there’s lots of other interesting examples.
For instance, we see a variety of small circle and eye detectors form
by piecing together early curves and circle detectors (conv2d2):
The circuit constructing a triangle detector.
The choice of which neurons in the previous layer to show is slightly cherrypicked for pedagogy.
The six neurons with the highest magnitude weights to the triangle are shown, plus one other neuron with slightly weaker weights.
(Left leaning edges have slightly higher weights than right ones, but it seemed more illustrative to show two of both.) Click on neurons to see the full weights.
However, in practice, these triangle detectors (and other angle units) seem to often just be used as multi-edge detectors downstream,
or in conjunction with many other units to detect convex boundaries.
The selected circuits discussed above only scratch the surface of the intricate structure in mixed3a.
Below, we provide a taxonomized overview of all of them:
This is a broad, not very well defined category for units that seem to look for simple local structures over a wide receptive field, including mixtures of colors. Many live in a branch consisting of a maxpool followed by a 1×1 conv, which structurally encourages this.Maxpool branches (ie. maxpool 5×5 stride 1 -> conv 1×1) have large receptive fields, but can’t control where in in their receptive field each feature they detect is, nor the relative position of these features. In early vision, this unstructured of feature detection makes them a good fit for textures.
Show all 30 neurons.
Collapse neurons.
These units look for one color in the center, and another (usually opposite) color surrounding it. They are typically much more sensitive to the center color than the surrounding one. In visual neuroscience, center-surround units are classically an extremely low-level feature, but we see them in the later parts of early vision. Compare to earlier Color Center-Surround (conv2d2) and later Color Center-Surround (mixed3b).
Show all 15 neurons.
Collapse neurons.
These units look for transitions from high-frequency texture to low-frequency. They are primarily used by boundary detectors (mixed3b) as an additional cue for a boundary between objects. (Larger scale high-low frequency detectors can be found in mixed4a (245, 93, 392, 301), but are not discussed in this article.)
A detailed article on these is forthcoming.
Show all 15 neurons.
Collapse neurons.
Show all 14 neurons.
Collapse neurons.
These units look for one color on one side of their receptive field, and another (usually opposite) color on the opposing side. They typically don’t care about the exact position or orientation of the transition. Compare to earlier color contrast (conv2d0, conv2d1, conv2d2) and later color contrast (mixed3b).
Show all 14 neurons.
Collapse neurons.
This is a broad, not very well defined category for center-surround units that detect a pattern or complex texture in their center.
Units used to detect extended lines, often further excited by different colors on each side. A few are highly combed line detectors that aren’t obviously such at first glance. The decision to include a unit was often decided by whether it seems to be used by downstream client units as a line detector.
Show all 14 neurons.
Collapse neurons.
Catch-all category for all other units.
Show all 12 neurons.
Collapse neurons.
This is broad, catch-all category for units that seem to look for repeating local patterns that seem more complex than textures.
Show all 11 neurons.
Collapse neurons.
These “black and white” detectors respond to absences of color. Prior to this, color detectors contrast to the opposite hue, but from this point on we’ll see many compare to the absence of color. See also BW circuit example and discussion.
Units that detect multiple lines, forming angles, triangles and squares. They generally respond to any of the individual lines, and more strongly to them together.
These units are not yet highly selective for fur (they also fire for other high-frequency patterns), but their primary use in the next layer is supporting fur detection. At the 224×224 image resolution, individual fur hairs are generally not detectable, but tufts of fur are. These units use Gabor textures to detect those tufts in different orientations. The also detect lower frequency edges or changes in lighting perpendicular to the tufts.
These units seem to respond to lines crossing or to lines diverging from a central point.
Low confidence organizational category.
These units detect a line ending or sharply turning. Often used in boundary detection and more complex shape detectors.
mixed3b
mixed3b straddles two levels of abstraction.
On the one hand, it has some quite sophisticated features that don’t really seem like they should be characterized as “early” or “low-level”: object boundary detectors, early head detectors, and more sophisticated part of shape detectors.
On the other hand, it also has many units that still feel quite low-level, such as color center-surround units.
Boundary detectors: One of the most striking transitions in mixed3b is the formation of boundary detectors.
When you first look at the feature visualizations and dataset examples,
you might think these are just another iteration of edge or curve detectors.
But they are in fact combining a variety of cues to detect boundaries and transitions between objects.
Perhaps the most important one is the high-low frequency detectors we saw develop at the previous layer.
Notice that it largely doesn’t care which direction the change in color or frequency is, just that there’s a change.
We sometimes find it useful to think about the “goal” of early vision.
Gradient descent will only create features if they are useful for features in later layers.
Which later features incentivized the creation of the features we see in early vision?
These boundary detectors seem to be the “goal” of the high-low frequency detectors (mixed3a) we saw in the previous layer.
Curve-based Features:
Another major theme in this layer is the emergence of more complex and specific shape detectors based on curves.
These include more sophisticated curves, circles, S-shapes, spirals, divots, and “evolutes”
(a term we’ve repurposed to describe units detecting curves facing away from the middle).
We’ll discuss these in detail in a forthcoming article on curve circuits, but they warrant mention here.
Conceptually, you can think of the weights as piecing together curve detectors as something like this:
Fur detectors: Another interesting (albeit, probably quite specific to the dog focus of ImageNet)
circuit is the implementation of “oriented fur detectors” which detect fur parting, like hair on one’s head.
They’re implemented by piecing together fur precursors (mixed3a) so that they converge in a particular way.
Again, these circuits only scratch the surface of mixed3b.
Since it’s a larger layer with lots of families, we’ll go through a couple particularly interesting and well understood families first:
These units use multiple cues to detect the boundaries of objects. They vary in orientation, detecting convex/concave/straight boundaries, and detecting artificial vs fur foregrounds. Cues they rely on include line detectors, high-low frequency detectors, and color contrast.
Show all 12 neurons.
Collapse neurons.
The tiny eye detectors, along with texture detectors for fur, hair and skin developed at the previous layer enable these early head detectors, which will continue to be refined in the next layer.
We don’t typically think of fur as an oriented feature, but it is. These units detect fur parting in various ways, much like how hair on your head parts.
Curve-like detectors for sharp corners or bumps.
Units detecting grid patterns.
Again, we continue to see eye detectors quite early in vision. Note that several of these detect larger eyes than the earlier eye detectors (mixed3a). In the next layer, we see much larger scale eye detectors again.
Simple shapes created by composing curves, such as spirals and S-curves.
Piece together curves in a circle or partial circle. Opposite of evolute.
Units detecting circles and curves without necesarily requiring spatial coherrence.
Weights appear to be two curve detectors added together. Likely best thought of as a polysemantic neuron.
Detects curves facing away from the middle. Opposite of circles. Term repurposed from mathematical evolutes which can sometimes be visually similar.
In addition to the above features, are also a lot of other features which don’t fall into such a neat categorization.
One frustrating issue is that mixed3b has many units that don’t have a simple low-level articulation, but also are not yet very specific to a high-level feature.
For example, there are units which seem to be developing towards detecting certain animal body parts, but still respond to many other stimuli as well and so are difficult to describe.
These units look for one color in the center, and another color surrounding it. These units likely have many subtleties about the range of hues, texture preferences, and interactions that similar neurons in earlier layers may not have. Note how many units detect the absence (or generic presence) of color, building off of the black and white detectors in mixed3a. Compare to earlier Color Center-Surround (conv2d2) and (Color Center-Surround mixed3a).
This is a broad, not very well defined category for units that seem to look for simple local structures over a wide receptive field, including mixtures of colors.
Units which respond to different colors on each side. These units look for one color in the center, and another color surrounding it. These units likely have many subtleties about the range of hues, texture preferences, and interactions that similar neurons in earlier layers may not have. Compare to earlier color contrast (conv2d0, conv2d1, conv2d2, mixed3a).
Show all 12 neurons.
Collapse neurons.
Units that detect one texture on one side and a different texture on the other.
Units which seem to detect fur but, unlike the oriented fur detectors, don’t seem to detect it parting in a particular way. Many of these seem to prefer a particular fur pattern.
Units which seem, to a significant extent, to detect a line. Many seem to have additional, more complex behavior.
Units detecting lines crossing or diverging from a center point. Some are early predecessors for 3D corner detection.
Units that detect boundary transitions on two sides, with a ‘foreground’ texture in the middle.
Boundary-related units we didn’t know what else to do with.
We don’t really understand these units.
Low confidence organizational category.
Units that seem to detect shiny, specular surfaces.
Conclusion
The goal of this essay was to give a high-level overview of our present understanding of early vision in InceptionV1.
Every single feature discussed in this article is a potential topic of deep investigation.
For example, are curve detectors really curve detectors? What types of curves do they fire for? How do they behave on various edge cases? How are they built?
Over the coming articles, we’ll do deep dives rigorously investigating these questions for a few features, starting with curves.
Our investigation into early vision has also left us with many broader open questions.
To what extent do these feature families reflect fundamental clusters in features,
versus a taxonomy that might be helpful to humans but is ultimately somewhat arbitrary?
Is there a better taxonomy, or another way to understand the space of features?
Why do features often seem to form in families?
To what extent do the same features families form across models?
Is there a “periodic table of low-level visual features”, in some sense?
To what extent do later features admit a similar taxonomy?
We think these could be interesting questions for future work.
.Organize the content with appropriate headings and subheadings ( h2, h3, h4, h5, h6). Include conclusion section and FAQs section with Proper questions and answers at the end. do not include the title. it must return only article i dont want any extra information or introductory text with article e.g: ” Here is rewritten article:” or “Here is the rewritten content:”[/gpt3]
Black Friday Monitor Deals: Philips Evnia 34M2C8600 Steals the Show
A Monstrous Deal
Black Friday monitor deals can be overwhelming, with numerous monitors and price points to choose from. However, we’ve found a deal that’s hard to ignore – the Philips Evnia 34M2C8600, now available at an unbeatable price.
The Deal
Get the Philips Evnia 34M2C8600 for $599 on Amazon, a stunning 53% discount from its launch price of $1,299.99. This elegant ultrawide OLED monitor is not only a remarkable value but also a favourite among reviewers, including yours truly.
Why We Love It
This exceptional monitor boasts a 34-inch curved display, with a 3440 x 1440 resolution, 120Hz refresh rate, and HDR10+ support. Its sleek design and impressive performance make it an ideal choice for content creators, gamers, and anyone seeking a premium monitor experience.
Other Monitors to Consider
We understand that a 34-inch curved monitor might not be for everyone. Our team has reviewed and compared numerous monitors, so you can refer to our list of the Best Monitors for Creatives for more options.
Best Deal Today
Below, you’ll find the most competitive prices on the Philips Evnia 34M2C8600 across regions and worldwide, continuously updated by our deals widget.
Conclusion
This Black Friday, the Philips Evnia 34M2C8600 offers an unbeatable combination of performance, design, and value. If you’re in the market for an exceptional monitor, this is an opportunity you won’t want to miss.
Frequently Asked Questions
Q: Is the Philips Evnia 34M2C8600 only available on Amazon?
A: No, the Philips Evnia 34M2C8600 can be found on other e-commerce platforms and retailers. However, the price listed is exclusive to Amazon during this Black Friday sale.
Q: Can I still get the Philips Evnia 34M2C8600 at this price after Black Friday?
A: The deal is likely to expire after Black Friday, so it’s recommended to purchase the monitor by the end of the day on November 27 to secure the discounted price.
Generative AI tools such as ChatGPT and other assistants have the potential to improve teachers’ workloads, especially when it comes to lesson planning, grading, and personalized learning materials. However, knowing best practices is crucial to get the most out of the tech — and this new free AI course is meant to teach just that.
New Free AI Course for K-12 Educators
On Wednesday, OpenAI and Common Sense Media launched a free online course called "ChatGPT Foundations for K-12 Educators" to help teachers learn how to responsibly and effectively implement AI into their workflows.
Course Objectives
The course, which should take less than one hour to complete, covers the basics of AI, including topics such as data privacy implications, ethical usage, AI fundamentals, and ideas for how educators can leverage AI in their classrooms, according to the release. It was designed to be simple and accessible and will help educators understand the basics of AI beyond ChatGPT.
Pilot Results
The company shared that the course has already been piloted in a dozen school districts and has seen promising early results, with 98% of participants sharing that the course offered them new ideas or strategies they could apply to their work.
Availability
The course is available for free today on the Common Sense Media website for all educators and school districts. A ChatGPT Edu subscription is not required. This partnership marks the first public collaboration between the companies, with more to come, according to OpenAI.
Conclusion
The "ChatGPT Foundations for K-12 Educators" course is a valuable resource for teachers looking to improve their understanding of AI and its applications in the classroom. By providing educators with the knowledge and skills they need to effectively integrate AI into their workflows, this course has the potential to improve teaching and learning outcomes for students.
Frequently Asked Questions
Q: What is the purpose of the "ChatGPT Foundations for K-12 Educators" course?
A: The course is designed to help teachers learn how to responsibly and effectively implement AI into their workflows.
Q: How long does the course take to complete?
A: The course should take less than one hour to complete.
Q: Is the course available for free?
A: Yes, the course is available for free on the Common Sense Media website.
Q: Do I need a ChatGPT Edu subscription to take the course?
A: No, a ChatGPT Edu subscription is not required.
Q: What are the topics covered in the course?
A: The course covers the basics of AI, including data privacy implications, ethical usage, AI fundamentals, and ideas for how educators can leverage AI in their classrooms.
Support me on Patreon: https://www.patreon.com/sarikas
Introduction
As a content creator, I strive to provide high-quality and engaging content to my audience. However, creating content requires a significant amount of time, effort, and resources. To continue producing the best content possible, I need your support.
Why Patreon?
Patreon is a platform that allows creators to receive recurring support from their fans and patrons. By becoming a patron, you can help me continue to create content that you enjoy. In return, you’ll receive exclusive rewards and benefits that will enhance your experience.
What Can You Expect?
As a patron, you can expect:
* Exclusive access to behind-the-scenes content, including videos, photos, and blog posts
* Early access to new content, including articles, videos, and podcasts
* Personalized thank-you messages and shoutouts on social media
* Discounts on merchandise and other products
* The satisfaction of knowing that you’re supporting a creator who is passionate about their work
How Can You Support Me?
You can support me on Patreon by becoming a patron and contributing a monthly amount of your choice. You can also share my content with your friends and family, and leave comments and feedback to help me improve.
Conclusion
I want to thank you for considering supporting me on Patreon. Your contributions will help me to continue creating content that you enjoy, and I’m grateful for your support.
FAQs
Q: What is Patreon?
A: Patreon is a platform that allows creators to receive recurring support from their fans and patrons.
Q: What kind of content can I expect as a patron?
A: As a patron, you can expect exclusive access to behind-the-scenes content, early access to new content, personalized thank-you messages, and more.
Q: How much does it cost to support you on Patreon?
A: You can contribute any amount you choose, from $1 to $100 per month.
Q: Can I cancel my Patreon subscription at any time?
A: Yes, you can cancel your Patreon subscription at any time.
Under a Trump administration, the outcomes for 2025 appear inevitably bleaker. He’s promised to invest in an economy that is anti-woke, bolstering his cabinet with agitators—such as Brendan Carr, his choice for chair of the Federal Communications Commission—that have promised to end DEI. Project 2025, the 900-page conservative policy agenda Trump is likely to base much of his governing on, takes aim at organizations that employ “racial classifications and quotas” and pledges to rescind an executive order that calls for federal contractors to guarantee equal opportunity. (And big tech companies were already cutting DEI programs, even without the threats of a hostile president.)
AfroTech Conference
At the AfroTech conference, the vibe was optimistic, despite the uncertainty. The conference featured a recruiting expo, where big players like Netflix, American Express, Axon, Meta, Google, and Oracle were represented. The exhibition floor was buzzing with activity, and the atmosphere was electric.
Concerns about Job Security
However, beneath the surface, concerns about job security were palpable. Many attendees were worried about the impact of a potential Trump administration on their careers and the future of diversity and inclusion initiatives. Candace Madison, a first-time attendee, expressed her concerns: "I don’t think the election increased that, but with the election and DEI not being a priority, you do have to be more on your toes."
Networking and Uncertainty
Many attendees were focused on networking and building connections, aware that the job market would only become more competitive in the future. A graduate student completing her PhD in data science noted: "This is my eighth conference of the year… I’m doing my best networking, but I’m not getting much out of them."
Promising Outcomes
Despite the uncertainty, the conference had its share of success stories. A product engineer at a Fortune 50 company shared his story of how the conference had helped him land an internship and his current job. A high-ranking marketing executive described the conference as "a balm in Gilead."
Conclusion
The AfroTech conference was a testament to the resilience and determination of its attendees, who were focused on building a better future despite the uncertainty ahead. However, the looming threat of a Trump administration and its anti-woke agenda cast a shadow over the proceedings. As one attendee noted: "We’ve never seen what’s about to come."
FAQs
Q: What is the AfroTech conference?
A: The AfroTech conference is an annual event that brings together professionals and students in the tech industry, with a focus on diversity and inclusion.
Q: What was the mood at the conference?
A: The mood was optimistic, despite concerns about the future of diversity and inclusion initiatives under a potential Trump administration.
Q: What were the attendees concerned about?
A: Attendees were concerned about job security, the impact of a potential Trump administration on their careers, and the future of diversity and inclusion initiatives.
Q: What was the atmosphere like at the conference?
A: The atmosphere was electric, with a sense of excitement and optimism despite the uncertainty ahead.
Nvidia said its latest generation of chips are in “great shape” as the world’s most valuable company reported another strong quarter of revenue growth thanks to high demand for the infrastructure that has underpinned the artificial intelligence boom.
Revenue Growth
Revenue for the quarter to the end of October nearly doubled from a year ago, up 94 per cent to $35.1bn, according to a release on Wednesday. It was a slower pace of growth from the previous quarter but still well above analysts’ expectations for $33.25bn.
Blackwell Chips
Analysts have been watching closely to see how Nvidia’s new generation of chips, known as Blackwell, launched earlier this year, might affect short-term revenue growth, and whether the chip is encountering any technical issues as it is implemented at scale.
According to a recent report from The Information, the Blackwell chips have experienced problems with overheating in servers. The chip already faced production issues earlier this year.
CEO Comments
Asked about the report, Nvidia chief executive Jensen Huang said Nvidia was selling more Blackwell products than expected this quarter, exceeding its previous forecast for “several billion” dollars in sales for the fiscal year.
“We will deliver, this quarter, more Blackwells than we had previously estimated, and so the supply chain team is doing an incredible job working with our supply partners to increase Blackwell,” he said. “Blackwell is in great shape.”
Supply Chain Constraints
Constraints in Nvidia’s supply chain, including with its crucial manufacturing partner TSMC, have prompted analysts to question how fast the new chips can be brought online in data centres.
Financial Performance
Gross margins were 75 per cent, in line with consensus estimates. Adjusted net income was $20bn, while earnings per share was $0.78, exceeding analysts’ expectations.
Conclusion
Nvidia’s strong quarter is a testament to the company’s dominance in the AI market and its ability to deliver high-performance chips that meet the growing demand for data centre infrastructure. Despite some technical issues with its new Blackwell chips, Nvidia is confident that it can deliver on its revenue guidance and continue to grow its business.
FAQs
Q: What was Nvidia’s revenue growth for the quarter?
A: Nvidia’s revenue grew 94 per cent to $35.1bn for the quarter to the end of October.
Q: What are the concerns about Nvidia’s Blackwell chips?
A: Some analysts have raised concerns about overheating issues with the Blackwell chips in servers, as well as production issues earlier this year.
Q: What is Nvidia’s guidance for the current quarter?
A: Nvidia has guided for revenue of $37.5bn, plus or minus 2 per cent, for the current quarter.
Q: What is the significance of Nvidia’s earnings for the tech market?
A: Nvidia’s earnings are closely watched as a measure of the health of the overall tech market, with all the biggest tech companies making significant investments in AI.
Sure, Black Friday 2024 is technically next Friday, November 29, but Amazon’s calling it Black Friday this week, and it has some epic laptop deals to back that claim. Highlights include a record $250 off this year’s M3 MacBook Air with 16GB of memory.
Laptop Deals for General Home Use
I’ve picked through all the early Amazon Black Friday deals to pick out the biggest savings on laptops for different needs, from general home use to options for power users who need a machine that can handle demanding software. Here are some of the top deals:
Acer Aspire 3 Laptop – $230 (was $350)
HP Envy x360 Laptop – $280 (was $400)
Lenovo IdeaPad 330S Laptop – $250 (was $350)
Laptop Deals for Power Users
If you’re a power user who needs a laptop that can handle demanding software, here are some deals that might interest you:
Microsoft Surface Laptop 4 – $500 (was $700)
Asus ZenBook 14 Laptop – $300 (was $400)
Dell XPS 13 Laptop – $400 (was $600)
MacBook Deals
If you’re specifically looking for MacBooks, check out Amazon’s surprise early Black Friday MacBook Pro M4 deals. In the meantime, here are some other MacBook deals:
Apple MacBook Air M3 – $1,049 (was $1,299)
Apple MacBook Pro M2 – $1,099 (was $1,299)
Conclusion
Amazon’s Black Friday laptop deals are already live, and they’re offering some incredible savings on a range of laptops. Whether you’re looking for a laptop for general home use or something more powerful for demanding software, there’s something for everyone. Don’t miss out on these deals – they might not last long!
FAQs
Q: When is Black Friday 2024?
A: Black Friday 2024 is on November 29.
Q: Are these deals available in the UK?
A: Yes, some of these deals are already available in the UK, but prices may vary.
Q: Can I get a better deal on a laptop elsewhere?
A: It’s always worth checking prices at other retailers to make sure you’re getting the best deal. However, Amazon’s Black Friday deals are often competitive, and you may not find better prices elsewhere.
Q: Do I need to be an Amazon Prime member to get these deals?
A: No, you don’t need to be an Amazon Prime member to get these deals. However, Prime members may get additional benefits, such as free shipping and early access to deals.
Samsung Reveals Gauss2, the Second-Generation Proprietary AI Model
Samsung has announced Gauss2, the second-generation proprietary AI model designed to significantly enhance user experiences. The new model is a multimodal AI that can handle language, code, and images, and comes in three variants: Compact, Balanced, and Supreme.
Multimodal AI
Gauss2 builds upon last year’s internal generative AI model, showcasing advancements in handling language, code, and images. The Compact variant is optimized for environments with limited computing resources, ensuring effective performance even on-device. The Balanced variant strikes a balance among performance, speed, and efficiency, making it suitable for a variety of tasks. The Supreme variant incorporates Mixture of Experts (MoE) technology to minimize computational costs while maximizing efficiency and performance.
Customization and Deployment
Samsung’s in-house generative AI model allows for easier customization for specific applications. Gauss2 supports diverse productivity tasks, with widespread adoption among Samsung developers. The coding assistant ‘code.i’ – enhanced by Gauss2 – is utilized extensively within the DX Division and by international research teams, with up to 60% of developers now engaging regularly.
Future Plans
Going forward, Samsung aims to further boost internal productivity, improving services like code.i and enhancing the Gauss Portal’s natural language question-and-answer capabilities. Future functionalities will include multimodal operations, like chart analysis and image creation.
‘AI for All’
Samsung’s strategy, "AI for All," envisions widespread incorporation of AI-based services across its product lines. By combining AI with knowledge graph technology, Samsung anticipates delivering even greater personalization.
Conclusion
Samsung’s Gauss2 is a significant advancement in AI technology, designed to enhance user experiences across various applications. With its multimodal capabilities and customization options, Gauss2 has the potential to revolutionize the way we interact with technology.
Frequently Asked Questions
Q: What is Gauss2?
A: Gauss2 is Samsung’s second-generation proprietary AI model designed to enhance user experiences.
Q: What are the variants of Gauss2?
A: Gauss2 comes in three variants: Compact, Balanced, and Supreme.
Q: What is the purpose of Gauss2?
A: Gauss2 is designed to support diverse productivity tasks and enhance services like code.i and the Gauss Portal.
Q: What is Samsung’s strategy for AI?
A: Samsung’s strategy, "AI for All," envisions widespread incorporation of AI-based services across its product lines.
/cdn.vox-cdn.com/uploads/chorus_asset/file/25263502/STK_414_AI_B.jpg "Filing Supporting Minnesota Deepfake Law May Have Been Written by AI")
