Connect with us


Facebook Researcher’s New Algorithm Ushers New Paradigm Of Image Recognition



“VICReg could be used to model the dependencies between a video clip and the frame that comes after, therefore learning to predict the future in a video.”

Adrien Bardes, Facebook AI Research

Humans have an innate capability to identify objects in the wild, even from a blurred glimpse of the thing. We do this efficiently by remembering only high-level features that get the job done (identification) and ignoring the details unless required. In the context of deep learning algorithms that do object detection, contrastive learning explored the premise of representation learning to obtain a large picture instead of doing the heavy lifting by devouring pixel-level details. But, contrastive learning has its own limitations. 

According to Andrew Ng, pre-training methods can suffer from three common failings: generating an identical representation for different input examples (which leads to predicting the mean consistently in linear regression), generating dissimilar representations for examples that humans find similar (for instance, the same object viewed from two angles), and generating redundant parts of a representation (say, multiple vectors that represent two eyes in a photo of a face). The problems of representation learning, wrote Andrew Ng, boil down to variance, invariance, and covariance issues. 

Register for our upcoming AI Conference>>

Also Read: What is Contrastive Learning

Andrew Ng’s observations are a reference to a new self-supervised algorithm released by the researchers at Facebook AI, PSL Research University, and New York University, along with Turing award recipient Yann Lecun introduced called Variance-Invariance-Covariance Regularization (VICReg), which builds on Lecun’s own Barlow Twins method.

(Image credits: Bardes et al.,)

The researchers designed VICReg (Variance-Invariance-Covariance Regularization) to avoid the collapse problem, which is handled more inefficiently in the case of contrastive methods. They do this by introducing a simple regularisation term on the variance of the embeddings along each dimension individually and combining the variance term with a decorrelation mechanism based on redundancy reduction and covariance regularisation. The authors state that VICReg is performed on par with several state-of-the-art methods.

See also  Facebook seeks replacement for former policy chief Ankhi Das

VICReg is a simple approach to self-supervised image representation learning, and its objectives are as follows:

  • Learn invariance to different views with an invariance term.
  • Avoid collapse of the representations with a variance regularisation term.
  • Spread the information throughout the different dimensions of the representations with a covariance regularisation term. 

The results show that VICReg performs on par with state-of-the-art methods and ushers a new paradigm of non-contrastive self-supervised learning. 

What Authors Had To Say

VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning.

By Adrien Bardes, Jean Ponce, and yours truly.

Insanely simple and effective method for self-supervised training of joint-embedding architectures (e.g. Siamese nets).


— Yann LeCun (@ylecun) May 12, 2021

Talking to Analytics India Magazine about VICReg’s significance, the lead author, Adrien Bardes, who is also a resident PhD student at Facebook AI Research, Paris, said that self-supervised representation learning is a learning paradigm that aims to learn meaningful representations of some unlabelled data. Recent approaches rely on Siamese networks and maximise the similarity between two augmented views of the same input. A trivial solution is for the network to output constant vectors, known as the collapse problem. VICReg is a new algorithm based on siamese networks but aims to prevent a collapse by regularising the variance and covariance of the network outputs. It achieves state-of-the-art results in several computer vision benchmarks while being a straightforward and interpretable approach.

When asked about how VICReg addresses shortcomings of contrastive learning methods, Bardes explained that contrastive learning methods are based on a simple principle. They make the inputs that should encode similar information close to each other in the embedding space and prevent a collapse by pushing apart the inputs that should encode dissimilar information. This process requires the mining of a massive amount of negative pairs, pairs of distinct inputs. Recent contrastive approaches for self-supervised learning have different strategies for mining these negative pairs; they can sample them from a memory bank, as in MoCo, or sample them from the current batch, as in SimCLR, which in both cases is costly in time or memory. VICReg, on the other hand, does not require these negative pairs; it implicitly prevents a collapse by enforcing the representations to be different from each other without making any direct comparison between different examples. It, therefore, does not require the memory bank of MoCo and works with much smaller batch sizes than SimCLR.

See also  Lindani Myeni's wife's heartbreaking Facebook tribute

For Bardes, self-supervised learning is probably the most exciting topic in machine learning research. Annotating data is a very expansive process performed by humans who have biases and can make mistakes. It is, therefore, impossible to annotate the vast amount of data available today, for example, medical or astronomical data and images and videos on the Internet. Training models that leverage all these data can only be done using self-supervised learning. This is one of the motivations behind the development of VICReg.

Bardes believes that VICReg is applicable in any scenario where one wants to model the relationships within a data set. It can be used with any kind of data, images, videos, text, audio, or proteins. For example, you could use it to model the dependencies between a video clip and the frame after, therefore learning to predict the future in a video. Another example would be to understand the relationship between the graph of a molecule and its image seen from a microscope.

“We are at the early stages of the development of self-supervised learning. Shifting from contrastive methods to non-contrastive methods is the first step towards more practical algorithms. Current approaches rely on hand-craft data augmentations that can be viewed as a kind of supervision. The next step will probably be to get rid of these augmentations. Another promising direction consists in handling the uncertainties in modelling the data. Current methods are mostly deterministic and always model the same relation between two inputs. For example, if we go back to the frame prediction example, current methods would only model the possible future for a video clip. Future approaches will probably use latent variables that model the space of possible predictions,” concluded Bardes.

See also  Looking for ways to deactivate your Facebook account on an iPhone? Check here - Information News

Join Our Discord Server. Be part of an engaging online community. Join Here.

Subscribe to our Newsletter

Get the latest updates and relevant offers by sharing your email.

Read More


Facebook Adds New Trend Insights in Creator Studio, Which Could Help Shape Your Posting Strategy




en flag
sv flag

Facebook’s looking to provide more content insight within Creator Studio with the rollout of a new ‘Inspiration Hub’ element, which highlights trending content and hashtags within categories related to your business Page.

Facebook Inspiration Hub

As you can see in these screenshots, posted by social media expert Matt Navarra, when it becomes available to you, you’ll be able to access the new Inspiration Hub from the Home tab in Creator Studio.

At the right side of the screen, you can see the first of the new insights, with trending hashtags and videos from the last 24 hours, posted by Pages similar to yours, displayed above a ‘See more’ prompt.

When you tap through to the new hub, you’ll have a range of additional filters to check out trending content from across Facebook, including Page category, content type, region, and more.

Facebook Inspiration Hub

That could be hugely valuable in learning what Facebook users are responding to, and what people within your target market are engaging with in the app.

The Hub also includes insights into trending hashtags, within your chosen timeframe, which may further assist in tapping into trending discussions.

Facebook Inspiration Hub

How valuable hashtags are on Facebook is still up for debate, but you’ll also note that you can filter the displayed results by platform, so you can additionally display Instagram hashtag trends as well, which could be very valuable in maximizing your reach.

Much of this type of info has been available within CrowdTangle, Facebook’s analytics platform for journalists, for some time, but not everyone can access CrowdTangle data, which could make this an even more valuable proposition for many marketers.

See also  Facebook hackers target small business owners to scam money for ads

Of course, overall performance really relates to your own creative, and thinking through the action that you want your audience to take when reading your posts. But in terms of detecting new content trends, including hashtag usage, caption length, videos versus image posts, and more, there’s a lot that could be gleaned from these tools and filters.

It’s a significant analytics addition – we’ve asked Facebook for more info on the rollout of the new option, and whether it’s already beyond test mode, etc. We’ll update this post if/when we hear back.

Continue Reading


Meta Updates Policy on Cryptocurrency Ads, Opening the Door to More Crypto Promotions in its Apps




en flag
sv flag

With cryptocurrencies gaining momentum, in line with the broader Web 3.0 push, Meta has today announced an update to its ad policies around cryptocurrencies, which will open the door to more crypto advertisers on its platforms.

As per Meta:

Starting today, we’re updating our eligibility criteria for running ads about cryptocurrency on our platform by expanding the number of regulatory licenses we accept from three to 27. We are also making the list of eligible licenses publicly available on our policy page.”

Essentially, in order to run any crypto ads in Meta’s apps, that currency needs to adhere to regional licensing provisions, which vary by nation. With crypto becoming more accepted, Meta’s now looking to enable more crypto companies to publish ads on its platform, which will provide expanded opportunity for recognized crypto providers to promote their products, while also enabling Meta to make more money from crypto ads.

“Previously, advertisers could submit an application and include information such as any licenses they obtained, whether they are traded on a public stock exchange, and other relevant public background on their business. However, over the years the cryptocurrency landscape has matured and stabilized and experienced an increase in government regulation, which has helped to set clearer responsibilities and expectations for the industry. Going forward, we will be moving away from using a variety of signals to confirm eligibility and instead requiring one of these 27 licenses.”

Is that a good move? Well, as Meta notes, the crypto marketplace is maturing, and there’s now much wider recognition of cryptocurrencies as a legitimate form of payment. But they’re also not supported by most local financial regulators, which reduced transaction protection and oversight, which also brings a level of risk in such process.

See also  Twitter tries reinventing itself, Facebook CPMs, and US newspaper readership

But then again, all crypto providers are required to clearly outline any such risks, and most also highlight the ongoing market volatility in the space. This expanded level of overall transparency means that most people who are investing in crypto have at least some awareness of these elements, which likely does diminish the risk factor in such promotions within Meta’s apps.

But as crypto adoption continues to expand, more of these risks will become apparent, and while much of the crypto community is built on good faith, and a sense of community around building something new, there are questions as to how much that can hold at scale, and what that will then mean for evolving scams and criminal activity, especially as more vulnerable investors are brought into the mix.

Broader promotional capacity through Meta’s apps will certainly help to boost exposure in this respect – though again, the relative risk factors are lessened by expanded regulatory oversight outside of the company.

You can read more about Meta’s expanded crypto ad regulations here.

Continue Reading


Meta Outlines Evolving Safety Measures in Messaging as it Seeks to Allay Fears Around the Expansion of E2E Encryption




en flag
sv flag

Amid rising concern about Meta’s move to roll out end-to-end encryption by default to all of its messaging apps, Meta’s Global Head of Safety Antigone Davis has today sought to provide a level of reassurance that Meta is indeed aware of the risks and dangers that such protection can pose, and that it is building safeguards into its processes to protect against potential misuse.

Though the measures outlined don’t exactly address all the issues raised by analysts and safety groups around the world.

As a quick recap, back in 2019, Facebook announced its plan to merge the messaging functionalities of Messenger, Instagram and WhatsApp, which would then provide users with a universal inbox, with all of your message threads from each app accessible on either platform.

The idea is that this will simplify cross-connection, while also opening the door to more opportunities for brands to connect with users in the messaging tool of their choice – but it also, inherently, means that the data protection method for its messaging tools must rise to the level of WhatsApp, its most secure messaging platform, which already includes E2E encryption as the default.

Various child safety experts raised the alarm, and several months after Facebook’s initial announcement, representatives from the UK, US and Australian Governments sent an open letter to Facebook CEO Mark Zuckerberg requesting that the company abandon its integration plan.

Meta has pushed ahead, despite specific concerns that the expansion of encryption will see its messaging tools used by child trafficking and exploitation groups, and now, as it closes in on the next stage, Meta’s working to counter such claims, with Davis outlining six key elements which she believes will ensure safety within this push.

See also  Facebook seeks replacement for former policy chief Ankhi Das

Davis has explained the various measures that Meta has added on this front, including:

  • Detection tools to stop adults from repeatedly setting up new profiles in an attempt to connect minors that they don’t know
  • Safety notices in Messenger, which provide tips on spotting suspicious behavior
  • The capacity to filter messages with selected keywords on Instagram
  • More filtering options in chat requests to help avoid unwanted contact
  • Improved education prompts to help detect spammers and scammers in messages
  • New processes to make it easier to report potential harm, including an option to select “involves a child”, which will then prioritize the report for review and action

Meta messaging security options

Which are all good, all important steps in detection, while Davis also notes that its reporting process “decrypts portions of the conversation that were previously encrypted and unavailable to us so that we can take immediate action if violations are detected”.

That’ll no doubt raise an eyebrow or two among WhatsApp users – but the problem here is that, overall, the broader concern is that such protections will facilitate usage by criminal groups, and the reliance on self-reporting in this respect is not going to have any impact on these networks operating, at scale, under a more protected messaging framework within Meta’s app eco-system.

Governments have called for ‘backdoor access’ to break Meta’s encryption for investigations into such activity, which Meta says is both not possible and will not be built into its future framework. The elements outlined by Davis do little to address this specific need, and without the capacity to better detect such, it’s hard to see any of the groups opposed to Meta’s expanded encryption changing their stance, and accepting that the merging of all of the platform’s DM options will not also see a rise in criminal activity organized via the same apps.

See also  The Facebook and Instagram memes bringing Indian history (and independence struggle) back to life

Of course, the counterargument could be that encryption is already available on WhatsApp, and that criminal activity of this type can already be undertaken within WhatsApp alone. But with a combined user count of 3.58 billion people per month across its family of apps, that’s a significantly broader interconnection of people than WhatsApp’s 2 billion active users, which, arguably, could open the door to far more potential harm and danger in this respect.

Really, there’s no right answer here. Privacy advocates will argue that encryption should be the standard, and that more people are actually more protected, on balance, by enhanced security measures. But there is also an undeniable risk in shielding even more criminal groups from detection.

Either way, right now, Meta seems determined to push ahead with the plan, which will weld all of its messaging tools together, and also make it more difficult to break-up its network, if any antitrust decisions don’t go Meta’s way, and it’s potentially pressed to sell-off Instagram or WhatsApp as a result.

But expect more debate to be had, in more countries, as Meta continues to justify its decision, and regulatory and law enforcement groups seek more options to help maintain a level of accessibility for criminal investigations and detection.

Continue Reading