Mediaboss Marketing

The performance of deep neural networks (DNNs) relies heavily on their structures, and designing a good structure (aka architecture) tends to require extensive effort from human experts. The idea of an automatic structure-learning algorithm that can achieve performance on par with the best human-designed structures is thus increasingly appealing to machine learning researchers.

In the paper Learning Structures for Deep Neural Networks, a team from OneFlow and Microsoft explores unsupervised structure learning, leveraging the efficient coding principle, information theory and computational neuroscience to design a structure learning method that does not require labelled information and demonstrates empirically that larger entropy outputs in a deep neural network lead to better performance.

The researchers start with the assumption that the optimal structure of neural networks can be derived from the input features without labels. Their study probes whether it is possible to learn good DNN network structures from scratch in a fully automatic fashion, and what would be a principled way to reach this end.

The team references a principle borrowed from the biological nervous system domain the efficient coding principle which posits that a good brain structure forms an efficient internal representation of external environments. They apply the efficient coding principle to DNN architecture, proposing that the structure of a well-designed network should match the statistical structure of its input signals.

The efficient coding principle suggests that the mutual information between a models inputs and outputs should be maximized, and the team presents a solid Bayesian optimal classification theoretical foundation to support this. Specifically, they show that the top layer of any neural network (softmax linear classifier) and the independency between the nodes in the top hidden layer constitute a sufficient condition for making the softmax linear classifier act as a Bayesian optimal classifier. This theoretical foundation not only backs up the efficient coding principle, it also provides a way to determine the depth of a DNN.

The team then investigates how to leverage the efficient coding principle in the design of a structure-learning algorithm, and shows that sparse coding can implement the principle under the assumption of zero-peaked and heavy-tailed prior distributions. This suggests that an effective structure learning algorithm can be designed based on global group sparse coding.

The proposed structure-learning with sparse coding algorithm learns a structure layer by layer in a bottom-up manner. The raw features are at layer one, and given the predefined number of nodes in layer two, the algorithm will learn the connection between these two layers, and so on.

The researchers also describe how this proposed algorithm can learn inter-layer connections, handle invariance, and determine DNN depth. Finally, they conduct intensive experiments on the popular CIFAR-10 data set to evaluate the classification accuracies of their proposed structure learning method, the role of inter-layer connections, and the role of structure masks and network depth.

The results show that a learned-structure single-layer network achieves an accuracy of 63.0 percent, outperforming the single-layer baseline of 60.4 percent. In an inter-layer connection density evaluation experiment, the structures generated by the sparse coding approach outperform random structures, and at the same density level, always outperform the sparsifying-restricted Boltzmann machines (RBM) baseline. In the teams structure mask role evaluation, the structure prior provided by sparse coding is seen to improve performance. The network depth experiment meanwhile empirically justifies the proposed approach for determining DNN depth via coding efficiency.

Overall, the research proves the efficient coding principles effectiveness for unsupervised structure learning, and that the proposed global sparse coding-based structure-learning algorithms can achieve performance comparable with the best human-designed structures.

The paper Learning Structures for Deep Neural Networks is on arXiv.

Author: Hecate He |Editor: Michael Sarazen, Chain Zhang

We know you dont want to miss any news or research breakthroughs.Subscribe to our popular newsletterSynced Global AI Weeklyto get weekly AI updates.

Like Loading...

View post:
Microsoft & OneFlow Leverage the Efficient Coding Principle to Design Unsupervised DNN Structure-Learning That Outperforms Human-Designed...

Artificial intelligence (AI) is ushering a new era of technological innovation, paving the way for increased adoption across several industries ranging from healthcare to e-commerce. The novel capabilities offered by AI are empowering businesses to automate the different operations, making them faster and smarter, and enhancing the overall productivity. One of the most significant benefits facilitated by AI is the management of massive datasets, which enables organizations to handle the flow of information efficiently. Overall, AI is making significant headway in the business world, generating robust benefits for integrators and adopters.

Bringing Transparency

Today, firms across the world are incorporating AI-based systems to automate their business processes and enable their workforce to focus on more valuable tasks. However, the incorporation of AI technology is impeded by several challenges, and the biggest one is the lack of transparency. AI systems are often considered as a black box, which makes it challenging to pinpoint logical errors in the underlying algorithms. These challenges in data privacy, protection, and cybersecurity have introduced nuances into the field, making it imperative to develop explainable AI, which offers greater visibility and transparency

Balancing Stability with Innovation

To make the most of AI, enterprises should also focus on incorporating the right tools, talent, and culture. While a myriad of different AI solutions has permeated the marketplace today, businesses still lack the expertise to identify which solution aligns with their organizational goals. This is where having the right partner to help you through every stage of the integration process makes all the difference. These partnerships should be compatible with the values, goals, and strategies of an organization. It is advisable to consider the risk factors and conduct an impact analysis on how the partnership will help drive business growth. For instance, I often collaborate with numerous partners, many of which possess robust and proven solutions. We also work with nascent companies that have a more novel approach to balance the risk versus the impact of the AI integrations.

When choosing partners, businesses should have a firm knowledge of the different areas of improvement within their organization. Successful collaboration relies on minimizing the risks and maximizing the benefits. By narrowing the AI integration to specific processes that require immediate upgrading, the overall workflow can be streamlined seamlessly. Once the relevant areas have been identified, the tech teams can decide on collaborations across the organizational line. Often, enterprises fail to see beyond the hype of AI products and rush into piloting several things without a clear vision of the end result. In such cases, businesses cannot derive the expected value from the solutions. Hence, it is crucial to have a specific goal in mind when integrating AI technology.

Augmenting AI with Robust Leadership

Along with having an AI-first mindset, businesses should be swift and versatile when executing on the AI technology. As a practitioner in this area for several years, I have witnessed a steady rise of interest in data science. One does not need a doctorate to be a data scientist. Perfection comes with practice. My advice to budding professionals is to follow their passion as they have a myriad of resources at their disposal, including journals, and coding courses. As for moving up the leadership ladder, practitioners must take bold decisions and get involved in the community. It also pays to take part in cooperative projects and contribute to the academic community by publishing papers, attending conferences, giving talks, and supporting the cause. Hence, as a leader, it is vital to adapt and change according to the trends, while also focusing on the key areas that need improvement.

Read more from the original source:
Explainable AI And The Future Of Machine Learning - CIO Applications

Last month, the U.S. Supreme Court held that police may not enter a private home without a search warrant to perform a community caretaking function. The Justices were unanimous in reaching this conclusion, though several concurred to clarify their points of view. One could understand the Justices as manifesting an unusual respect for privacy. I would instead read the decision as suggesting a somewhat different and more radical shift. The message of Defund the Police appears to have gotten through to our Supreme Court, perhaps without their fully realizing it.

In Caniglia v. Strom, the story began with Edward Caniglia placing a handgun on his dining room table and asking his wife to shoot him. His wife left the house and spent the night at a hotel, but she became worried the next morning when she was unable to reach her husband by phone. She called the police, and they accompanied her back to her house, where she found her husband on the porch, apparently unharmed. The police encouraged him to go for a psychiatric evaluation, which he agreed to do if they promised not to seize his guns from inside the house. Once Caniglia had left, however, the police entered the home and seized his weapons. Caniglia later sued the police for violating his Fourth Amendment rights when they entered the house and seized his property without a warrant and without consent.

The police and many court-watchers had assumed prior to this case that when police want to check on someones wellbeing, they are performing a community caretaking function and could legally enter the home without a warrant. In one Supreme Court case, Cady v. Dombrowski, police were able to look around inside a car that had been in an accident, under a community caretaking rationale.

Unlike other exceptions to the warrant requirement, however, it is unclear what the warrant application would have had to say in Caniglias case if police had first sought one. His wife was not reporting a crime, owning firearms is not necessarily criminal, and having been suicidally depressed violates no criminal law such that police might need to look for evidence of it. Once Caniglia had left for a psychiatric evaluation, it is hard to know what sort of probable cause the police had to authorize them to obtain a warrant for entering the home and seizing the mans guns. If a person is a danger to himself, then he might be subject to civil commitment, under Addington v. Texas, if the government can prove his mental illness and his danger to himself by clear and convincing evidence. But even attempted suicide is no longer a crime, as it once was, so the Fourth Amendment might limit what police could do at the investigative phase.

The police officers community caretaking function, at least as a broad category, falls outside the scope of typical police officer activity of the sort that rests on probable cause, with or without a warrant. It is the kind of conduct unrelated to conventional or core law enforcement responsibilities. And that is precisely what many people mean to remove from police jurisdiction and to invest in nonviolent alternatives when they press for defunding the police. They want the police to stop performing functions unrelated to criminal justice and to law and order. They argue that we should not be paying police officers to force themselves into peoples lives to address non-criminal problems.

Several Justices wrote separately to qualify the majority opinion and say that police may well have the authority, consistent with the Fourth Amendment, to enter homes without a warrant if a true health emergency presents itself. Justice Alito, for example, maintained that permissible community caretaking by the police could include conducting a search or seizure for the purpose of preventing a person from committing suicide. And if neighbors of a resident worry that the resident is in urgent need of medical attention and cannot summon help, Justice Alito strongly implied that he would view a warrantless entry to check on the residents condition as legitimate and lawful under the Fourth Amendment. Justice Kavanaugh, going further, said that police officers may enter a home without a warrant in circumstances where they are reasonably trying to prevent a potential suicide or help an elderly person who has been out of contact and may have fallen and suffered a serious injury.

Notably, however, the various concurring opinions limiting the impact of the unanimous majority amounted to only four votes: the Chief Justice, Justice Breyer, Justice Alito, and Justice Kavanaugh. The reservation of some community caretaking for policesuicidality and checking on a neighbor specificallythus remains uncertain. In keeping with this uncertainty, just yesterday, the Supreme Court granted, vacated, and remanded (GVRed) in Sanders v. United States, in light of Caniglia. GVRs generally indicate that a recent decision plainly dictates the result in the remanded case. Accordingly, yesterdays GVR suggests that Caniglia and its potential to defund the police is not a fact-bound and narrow one-off. To be sure, Justice Kavanaugh wrote separately, concurring in the GVR, to suggest ways in which the lower court might resolve the remanded case in favor of the governments authority to enter without a warrant. But this resistance may signal its own opposite as powerfully as the Courts failure to either hear the new case on the merits or simply deny certiorari without comment.

Returning to Caniglia itself, what surprised me as a reader was the assumption by the various concurrences that if police, an armed and quasi-military body, cannot enter a home to help a resident, then no one will be able to assist the resident. It does not even seem to occur to any of the four who wrote or joined a separate concurrence that perhaps a social worker, someone experienced in investigating reports of child neglect and perhaps trained in mental health or geriatric needs, could offer superior intervention to whats on offer from a police officer trained primarily to investigate crime. Defunding the police here could mean a reallocation of responsibilities so that more qualified people enter homes to help someone who is not a criminal suspect or a suspected victim of crime. The unanimous majority, coupled with only a minority wanting to reserve these jobs for law enforcement, could require such reallocation.

If we consider some of the cases in which police have used deadly force on unarmed civilians, we see the cost of sending cops to where their expertise does not extend. A study called Overlooked in the Undercounted: the Role of Mental Illness in Fatal Law Enforcement Encounters explains that although untreated mentally ill adults make up only 2% of the population, they make up a quarter of all fatal police encounters and are also sixteen times more likely to be killed during a police encounter than other civilians approached or stopped by police. It would seem imperative that someone trained in helping the mentally ill be the one involved in providing such help.

Reflecting such thinking, the city of Austin, Texas, created a mental health option for 9-1-1 callers. Such a move might save lives, in light of what too often happens when armed police show up at a home after a 9-1-1 call by the family member of a mentally ill individual. People trained in interacting with those who have mental disorders are unlikely to use (or even to have the option of using) deadly force against a person suffering a psychotic episode that precludes his putting his hands in the air or otherwise demonstrating submission to police.

Even if some of the same people (i.e., police officers) are involved in responding to mental health-related emergency calls, those peoplein keeping with sociologist Erving Goffmans dramaturgical theory of behaviorwill occupy a different role and thus behave differently in the presence of challenging behavior. Goffman viewed people as more like actors on a stage than stable characters who always behave the same way in every situation. Austin, Texas, may provide a natural experiment for this theory because many of the people who will be responding to 9-1-1 mental health calls are the same police officers who might have once responded to such calls in the role of law-and-order official. The difference is that with the new option, the officers will arrive on the scene wearing their social worker hats rather than reaching for their guns.

With the U.S. Supreme Courts decision denying police the authority to perform a community caretaking function by entering a home without a warrant to seize a potentially suicidal persons weapons, we might see cities besides Austin reassigning non-criminal police work to people wearing the mental health worker hat. In effectively defunding the police, the Court will eventually have to clarify what non-police actors must do to comply with the Fourth Amendment as they carry out community caretaking functions. A warrant, based upon probable cause to believe one will find evidence of crime or criminals in a private home, seems ill-suited to such functions.

Let us hope that either the Supreme Court or cities like Austin continue to contract the category of situations in which police have the authority to enter a home, guns drawn. It would be a pleasant surprise if the Roberts Court led the way in defunding the police.

See more here:
The U.S. Supreme Court Takes a Step toward Defunding the Police - Justia Verdict

Here we go again. The corporate influence on elections and public policies is again in the news.

The 2021 version centers on major corporations speaking out against proposed and passed voter-suppression laws in Georgia and elsewhere.

When some corporations suspended contributions to Republican politicians who voted against certifying Bidens election, the GOP expressed outrage that corporations dared to take political positions against them.

The GOP simultaneously bemoaned the prospect that the tsunami of corporate campaign contributions might be cut off or reduced. Just shut up and donate, was the message from Mitch McConnell.

My warning to corporate America is to stay out of politics, but Im not talking about political contributions, he stated in response to Coca-Cola, Delta Airlines and other national and transnational corporations that criticized new voter-suppression laws across the country. This is a brand new same old story.

var vf_ad_container = document.getElementById(vf_div_id);vf_ad_container.innerHTML = '

';AdBridg.cmd.push(function() {var vf_gpt_slot = AdBridg.defineSlot('/5195/NCPC_TheUnion/ROS/opinion', [300,250], vf_div_id).setTargeting('slot', vf_div_id);var vf_size_mapping = AdBridg.sizeMapping().addSize([320,50]).build();AdBridg.useSizeMapping(vf_gpt_slot, vf_size_mapping);AdBridg.display('ad-big-box4');AdBridg.serve();})

var r1 = vf_ad_container.closest('.vf-promo');if (typeof r1 !== undefined) r1.style.display = "block";

});

});

Remember the Supreme Courts 2010 Citizens United decision, which entrenched the notion that money is First Amendment protected free speech? This ruling allowed individuals and corporations to legally spend vast amounts of money in elections, much of it on attack ads targeting candidates on issues the funders oppose.

This decision incensed the public, which feared that the megaphone of mega money would drown out the voices of the vast majority of individuals who dont make big donations.

Here we are again. The public is justly critical of the massive hypocrisy displayed by Mitch and his GOP cohorts who welcome the cascade of corporate and CEO cash, while simultaneously expressing indignation when their corporate donors express political free speech opinions the GOP does not like.

Recall that the GOP did not attack corporate so-called free speech rights when corporations released press statements supporting the 2017 tax cut legislation for corporations and the wealthy.

As with Citizens United, the hypocritical application of the court invented so-called corporate free speech rights invites us to expose and abolish the totality and illegitimacy of anti-democratic corporate constitutional rights.

But even if corporate political free speech rights is abolished, corporate entities will still possess other century-old Supreme Court-invented anti-democratic constitutional rights that shield them from public accountability, and gives them undue political power and ensures that corporate profits preempt the rights of individuals, communities and even the natural world.

Among these never-intended corporate constitutional rights:

First Amendment right not to speak used to overturn state laws requiring the labeling of a dangerous ingredient on food products and chemicals determined to cause cancer.

Fourth Amendment search and seizure rights used to prevent surprise inspections of corporate property, even routine inspections, removing the ability of inspectors to detect dangerous conditions (food contamination, dangerous working conditions) before they are temporarily removed or covered up.

Fifth Amendment takings rights used to overturn public regulations that protect private property from corporate actions. Fossil fuel corporations would undoubtedly challenge laws to keep fossil fuels in the ground to prevent impending climate collapse as a taking of corporate property without just compensation.

Fourteenth Amendment equal protection rights used to overturn laws providing protection to local businesses (and local economies) over chain stores, and due process rights used to overturn over 200 state and federal economic regulations.

The corporate hijacking of these constitutional amendments provide business corporations overwhelming unaccountable power to overturn democratically enacted laws at every level of government.

The solution is the We the People Amendment (HJR 48), recently introduced by Rep. Pramila Jayapal, D-Wash., with 56 House co-sponsors. The constitutional amendment would abolish both the money equals speech and corporate constitutional rights doctrines.

Its a response to the fundamental truth that solutions must be equivalent in scale to problems, which in this case are the massive corrupting influences of money in elections from the super rich and corporate control over so many aspects of our lives, as well as plundering of the natural world.

Corporations shouldnt be allowed to meddle in elections. Its up to human persons to fundamentally define what our legal creations can and cannot do. Enactment of the We the People Amendment is a necessary and urgent step toward authentic democracy for all people. Support HJR 48 by calling your congressman and senator today!

Lorraine Reich is a member of Move to Amend, Nevada County.

Continue reading here:
Lorraine Reich: It's more than corporate money in elections - The Union of Grass Valley

This article is part of the Policing and Technology Project, a collaboration between Future Tense and the Tech, Law, & Security Program at American University Washington College of Law that examines the relationship between law enforcement, police reform, and technology.

The police show up to your house. Its the middle of the night, you are disoriented, and they want to know where you were earlier in the day. You have no idea at that moment that your ex-girlfriend was found dead, and some of your fingerprints were found at her housebut you do know you have the right to remain silent. Until the cops bring out the headset.

One of the hallmarks of the U.S. Constitution is the enumerated right of citizens to not be coerced into self-incrimination or be allowed to take the Fifth. But new technologies may one day be able to read your mind to varying degrees, rendering your decision to stay silent moot. While current devices merely collect data such as brain activity, labs are working on revolutionary devices that can record thoughts or allow for telepathic communication. They may be years or decades away, but they are worth thinking about now. As the courts seem to be moving toward allowing more and more personal data to be used as evidence. This data may eventually be both a window into mind and a side-step to Fifth Amendment protection.

Brain-computer interface devices are poised to become an integral treatment for diseases of the nervous system, by restoring brain function, mapping the brain, and enhancing cognitive function. These devices function through direct communication between the signals from a persons brain and an external computer. Some BCI devices are already on the market, though they are more quotidian: Muse, for instance, offers a wearable EEG device to aid in meditation, already have devices on market. Elon Musks Neuralink and Synchrons Stentrode, which both aim to return motor function to patients with neuromuscular conditions such as paralysis, are working on more invasive interventions, which would require surgery for implantation. Neuralink is designing a robot-driven brain surgery, while Synchron will be implanted via the patients blood vessels. Both have received breakthrough status from the FDA, meaning they will have an accelerated regulatory review process.

Eventually, BCIs could allow paralyzed people to walk, use their arms to get dressed, or communicate verbally. Should that happen, these devices will have unprecedented access to the human mind and even an individuals thoughts themselves. Though there are lots of ifs here and neural data is very noisy and hard to decode, studies have shown that synthetic speech can be generated from brain recordings.

And we have seen in the past, medical advancements can make their way into the criminal justice system, battering the boundaries of the Fifth Amendment. While it is easy to recognize verbal self-incrimination, it quickly becomes more complex once your mouth is no longer forming words. For example, if the police pull you over and ask whether you have been drinking, you may invoke the Fifth Amendment and decline to answer. However, the Fifth Amendment does not protect you from submitting to a field sobriety test and blood samples, even though they are ostensibly incriminating information gathered from the suspect. In some states, declining such a test can be treated as admission of guilt. This line of reasoning has now been applied to our cellphones and the ways they can be unlocked. For example, though you cannot be compelled to offer up your password to open your phone in most of the U.S., in many jurisdictions you may be forced to use your fingerprint or facial scan to do so. Why? Courts have reasoned that this is no different than a blood sample or left-behind fingerprint. Here, this becomes a simple reduction to what you know versus what you have. That distinction is likely to blur as technology develops.

On one hand, forcing a person to unlock a cellphone with a fingerprint seems vastly more invasive than finding a forgotten fingerprint theyve left behind. However, if we dont allow room for entry into the device in some fashion, that will only precipitate the creation of powerful technologies to access locked devices, which criminals may be able to use as easily as law enforcement. After the San Bernardino shootings in 2015, the FBI had a warrant to enter the deceased suspects iPhonebut no method of entry, as they had no passcode. Apple intentionally has not developed backdoor entry into its devices, which wipe themselves after 10 failed login attempts. Here, if not for federal hackers, this lawful evidence would have been lost forever. This ability to use the Fifth Amendment to intentionally hamstring the fourth amendment is also problematic.

To deal with this concern, the Supreme Court has developed a foregone conclusion test to apply to such cases. It says that if the state can demonstrate that it already knows what is on the device, it may compel the owner to provide the password or other means of entry. Unfortunately, this standard is vague. Courts vary in how narrowly or broadly they apply it, leaving behind disjointed rulings. These disjointed opinions are often the outcome of cases dealing with both the Fifth and Fourth Amendments, search and seizure.The Supreme Court can choose to hear cases that resulted in split decisions, but when the chance came to offer clarity on these questions in Jones v. Massachusetts in 2019, it refused to hear the case, so it will be some time before we have any resolution.

But the problem goes even deeper, as the actual invasion into the mind may not be necessary if the data is being collected in real time by a third party. In many instances this kind of medical data as fallen outside of self-incrimination discussions all together, being easily categorized as evidence and not testimony. Recently, a judge ruled pacemaker data admissible to demonstrate the defendants heart rate at time of a crime. The court reasoned that there was far more sensitive information than heart rate in the human body. However, as the heart rate is controlled by a nervous system response, one could also argue this is a rough look into the mind of the defendant. Is your nervous systems response at the time of a crime as simple as a left behind fingerprint? The courts have not established a clear demarcation between the mind and the body, which will be paramount for dealing with issues surrounding BCIs and self-incrimination. Here a basic protection for data collected from thoughts would provide a safeguard for cognitive liberty. Otherwise, we are left with loopholes that allow the state to have access to our most personal thoughts and motivations, which seems starkly against the spirit of the Fifth Amendment.

Moreover, as these BCI devices will save data, perhaps the Fifth Amendment will become less relevantafter all, companies will be holding onto treasure troves of neural data available for search under the Fourth Amendment. If the data is stored on a third-party device, such as a health app, it is not protected by the Fourth Amendmentunder the third-party doctrine, once you voluntarily give your information to, say, a company that provides a service, you waive any expectation of privacy and as such the state may have access. This guideline is commonly applied to phone records, but is it the standard we want in place with complex neural devices?

We need to create some new privacy rules for the 21st century that catch up to technologies new ability to assess not just our bodies but our minds. We should update the third-party doctrine so that cognitive data has more protections. Courts will inevitably have to interpret and apply more tests based on it, but they need a floor to work from in order to create more coherent and united opinions. If we do nothing, the Fifth Amendment could be weakened until its inevitable death.

Future Tense is a partnership of Slate, New America, and Arizona State University that examines emerging technologies, public policy, and society.

See more here:
If Police Have Devices That Can Read Your Mind, How Does the Fifth Amendment Fit In? - Slate