Connect with us


Looking into the black box




Deep learning systems are revolutionizing technology around us, from voice recognition that pairs you with your phone to autonomous vehicles that are increasingly able to see and recognize obstacles ahead. But much of this success involves trial and error when it comes to the deep learning networks themselves. A group of MIT researchers recently reviewed their contributions to a better theoretical understanding of deep learning networks, providing direction for the field moving forward.

“Deep learning was in some ways an accidental discovery,” explains Tommy Poggio, investigator at the McGovern Institute for Brain Research, director of the Center for Brains, Minds, and Machines (CBMM), and the Eugene McDermott Professor in Brain and Cognitive Sciences. “We still do not understand why it works. A theoretical framework is taking form, and I believe that we are now close to a satisfactory theory. It is time to stand back and review recent insights.”

Climbing data mountains

Our current era is marked by a superabundance of data — data from inexpensive sensors of all types, text, the internet, and large amounts of genomic data being generated in the life sciences. Computers nowadays ingest these multidimensional datasets, creating a set of problems dubbed the “curse of dimensionality” by the late mathematician Richard Bellman.

One of these problems is that representing a smooth, high-dimensional function requires an astronomically large number of parameters. We know that deep neural networks are particularly good at learning how to represent, or approximate, such complex data, but why? Understanding why could potentially help advance deep learning applications.

“Deep learning is like electricity after Volta discovered the battery, but before Maxwell,” explains Poggio, who is the founding scientific advisor of The Core, MIT Quest for Intelligence, and an investigator in the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. “Useful applications were certainly possible after Volta, but it was Maxwell’s theory of electromagnetism, this deeper understanding that then opened the way to the radio, the TV, the radar, the transistor, the computers, and the internet.”

The theoretical treatment by Poggio, Andrzej Banburski, and Qianli Liao points to why deep learning might overcome data problems such as “the curse of dimensionality.” Their approach starts with the observation that many natural structures are hierarchical. To model the growth and development of a tree doesn’t require that we specify the location of every twig. Instead, a model can use local rules to drive branching hierarchically. The primate visual system appears to do something similar when processing complex data. When we look at natural images — including trees, cats, and faces — the brain successively integrates local image patches, then small collections of patches, and then collections of collections of patches. 

“The physical world is compositional — in other words, composed of many local physical interactions,” explains Qianli Liao, an author of the study, and a graduate student in the Department of Electrical Engineering and Computer Science and a member of the CBMM. “This goes beyond images. Language and our thoughts are compositional, and even our nervous system is compositional in terms of how neurons connect with each other. Our review explains theoretically why deep networks are so good at representing this complexity.”

The intuition is that a hierarchical neural network should be better at approximating a compositional function than a single “layer” of neurons, even if the total number of neurons is the same. The technical part of their work identifies what “better at approximating” means and proves that the intuition is correct.

Generalization puzzle

There is a second puzzle about what is sometimes called the unreasonable effectiveness of deep networks. Deep network models often have far more parameters than data to fit them, despite the mountains of data we produce these days. This situation ought to lead to what is called “overfitting,” where your current data fit the model well, but any new data fit the model terribly. This is dubbed poor generalization in conventional models. The conventional solution is to constrain some aspect of the fitting procedure. However, deep networks do not seem to require this constraint. Poggio and his colleagues prove that, in many cases, the process of training a deep network implicitly “regularizes” the solution, providing constraints.

The work has a number of implications going forward. Though deep learning is actively being applied in the world, this has so far occurred without a comprehensive underlying theory. A theory of deep learning that explains why and how deep networks work, and what their limitations are, will likely allow development of even much more powerful learning approaches.

“In the long term, the ability to develop and build better intelligent machines will be essential to any technology-based economy,” explains Poggio. “After all, even in its current — still highly imperfect — state, deep learning is impacting, or about to impact, just about every aspect of our society and life.”

Topics: McGovern Institute, Center for Brains Minds and Machines, Brain and cognitive sciences, Quest for Intelligence, Computer Science and Artificial Intelligence Laboratory (CSAIL), Electrical engineering and computer science (EECS), Artificial intelligence, MIT Schwarzman College of Computing, School of Science, School of Engineering, Research



WCLC: Xcovery’s Xalkori challenger shines in phase 3 lung cancer showdown




Xcovery pitted its ALK inhibitor against Pfizer’s Xalkori—the first-ever FDA-approved drug in its class—and came out on top. The drug, ensartinib, shrank tumors in three-quarters of previously untreated lung cancer patients, compared to Xalkori’s 67%, and staved off cancer for two years, more than double the time Xalkori logged.

The phase 3 study tested ensartinib as a firstline treatment against Xalkori (crizotinib) in 290 patients with ALK-positive non-small cell lung cancer. As of July 1, ensartinib had shrunk the tumors of 75% of patients, topping Xalkori’s 67% mark. And even though Xalkori did unusually well in the study, keeping cancer at bay for a median of 12.7 months, Xcovery’s drug trounced it with median progression-free survival of 25.8 months. The median overall survival—how long the drug extended patients’ lives—had not been reached.

“Interestingly, crizotinib overperformed in the study; the duration of response for patients on crizotinib was 27.3 months, which is much higher than what we’ve seen in other studies for patients on crizotinib,” said Leora Horn, M.D., director of the Thoracic Oncology Program and a professor at the Vanderbilt-Ingram Cancer Center, who presented the study.

Sponsored by Biotech Primer

September 2-3, 2020 Live, Online Course: Biopharma Revenue Forecasting that Drives Decision Making and Investments

Become fluent in the core elements of revenue forecasting including epidemiology, competitive assessments, market share assignment and pricing. Let Biotech Primer’s dynamic industry experts teach you how to assess the value of new therapies.

“It was the best performance ever done by crizotinib in any randomized study” in previously untreated patients, said Xcovery Chief Medical Officer Giovanni Selvaggi, a thoracic oncologist by training who previously worked on Novartis’ ALK inhibitor Zykadia. 

RELATED: Xcovery hires CEO, CMO to lead assault on ALK NSCLC market

Ensartinib also did better than Xalkori in a small group of patients whose disease had spread to the brain. All 11 patients whose brain tumors were large and defined enough to be seen on a scan saw their tumors shrink, Selvaggi said. Seven of the 11 patients (64%) had enough shrinkage to be considered responders, compared to just one-fifth of the 19 patients on Xalkori.

The most common side effect was a sunburn-like rash, “a new toxicity” for ALK inhibitors, Horn said.

“It’s a very benign rash that can be compared to a sunburn and it normally goes away with continued treatment and topical therapies,” Selvaggi said. Unlike the rash that comes with meds that target EGFR, which often gets worse the longer patients take the drug, patients taking ensartinib do not need to stop treatment to get rid of the rash, he added.

The study started enrolling patients who were tested for the ALK mutation in local laboratories but about 40 patients in, it switched over to recruiting those who were centrally tested at larger labs, Horn said. It did this to cut down on false positive results from local testing. Of the 43 patients who were locally tested, 11 were thought to be ALK-positive before central testing showed they were actually negative; seven received ensartinib and two got Xalkori, Horn said.

“Since we only target the ALK protein with this drug, if patients are ALK-negative, it would not be expected to slow tumor growth,” Selvaggi said.

Understandably, the false positive patients would skew the data, so the investigators also reported results for what they called a modified intent-to-treat population of only centrally tested patients. The progression-free survival for this group had not been reached by the data cutoff, meaning more than half of the patients did not see their cancer worsen by then. This figure stayed at 12.7 months for patients on crizotinib.

RELATED: Pfizer’s Lorbrena makes play for earlier lung cancer use with Xalkori-topping data

Up next? Discussing the data with regulators and eventually, a filing: “We are excited to bring an effective and safe new option to patients and their physicians and therefore we are planning to share the data with regulatory agencies as our next step,” Selvaggi said. It’s been a long time coming for the Scripps Florida spinout that made the Fierce 15 class of 2007.

The thing is, if ensartinib can ultimately win approval, it’ll have more drugs to battle than just Xalkori. Roche’s Alecensa, the class’ current sales leader, and Zykadia both topped Xalkori on their way to earning first-line approvals in 2017, and Pfizer’s Lorbrena—Xalkori’s follow-up med—earlier this week posted top-line results showing it could beat out its predecessor, too.


Continue Reading


NIH to fund new diagnostic tests for the emerging, severe childhood illness linked to COVID-19




While most children exposed to the novel coronavirus suffer only a mild infection, others may develop a rare but severe reaction that attacks several organs at once, requiring intensive rescue care.

To better diagnose this complication of COVID-19—known as Multisystem Inflammatory Syndrome in Children, or MIS-C—the National Institutes of Health launched a research funding program that will offer up to $20 million in grants over the next four years.

MIS-C can be fatal, and has been shown to affect the heart, lungs, kidneys, brain, skin and eyes. A study by the Centers for Disease Control and Prevention tracking a group of MIS-C patients—at a median age of eight years old—showed 80% required intensive care, with many receiving ventilator support and treatments for circulatory shock.

Sponsored by Biotech Primer

September 2-3, 2020 Live, Online Course: Biopharma Revenue Forecasting that Drives Decision Making and Investments

Become fluent in the core elements of revenue forecasting including epidemiology, competitive assessments, market share assignment and pricing. Let Biotech Primer’s dynamic industry experts teach you how to assess the value of new therapies.

Most had tested positive for COVID-19 by either molecular or antibody tests, but early signs and symptoms can range from almost none at all to those seen and unseen—such as fever and cough, or abdominal pain and diarrhea, as well as hidden inflammation of the heart’s coronary arteries. 

The NIH’s new effort aims to incorporate new diagnostics for the different genetic, immune, viral and environmental factors that may help predict a child’s chances of progressing to MIS-C.

“We urgently need methods to distinguish children at high risk for MIS-C from those unlikely to experience major ill effects from the virus, so that we can develop early interventions to improve their outcomes, ” said Diana Bianchi, director of the NIH’s Eunice Kennedy Shriver National Institute of Child Health and Human Development.

RELATED: NIH picks seven COVID-19 diagnostic tests in ‘Shark Tank’ competition, unlocking $248.7M in scale-up funding

The institute’s project—dubbed PreVAIL kIds, for Predicting Viral-Associated Inflammatory Disease Severity in Children with Laboratory Diagnostics and Artificial Intelligence—dovetails with the NIH’s “Shark Tank-like” Rapid Acceleration of Diagnostics initiative, which recently advanced seven projects to its final phases.

PreVAIL kIds will support studies to evaluate genes and other pediatric biomarkers, and projects using machine learning techniques to help develop new tests, by gathering funds from multiple NIH institutes and centers focused on related organs, diseases and determinants of health.


Continue Reading


Biogen nabs speedy FDA review for controversial Alzheimer’s drug




Biogen’s aducanumab is inching closer to an FDA decision. The Big Biotech, along with partner Eisai, announced on Friday that the FDA accepted its regulatory submission for aducanumab, its once-failed Alzheimer’s drug—with priority review to boot.

The agency expects to decide the fate of the treatment by March 7, 2021. Along the way, it will hold an advisory committee meeting. It has not set a date for the meeting, but Jefferies analyst Michael Yee expects it sometime in the first quarter of 2021.

How the FDA rules on aducanumab will show how far the FDA and its commissioner, Stephen Hahn, M.D., are willing to diverge from its established approval standards. Under U.S. law, companies need to show “substantial” evidence of effectiveness to win approval.

Sponsored by Biotech Primer

September 2-3, 2020 Live, Online Course: Biopharma Revenue Forecasting that Drives Decision Making and Investments

Become fluent in the core elements of revenue forecasting including epidemiology, competitive assessments, market share assignment and pricing. Let Biotech Primer’s dynamic industry experts teach you how to assess the value of new therapies.

Aducanumab has had a bumpy ride, failing a futility analysis in March 2019, which led Biogen to axe its phase 3 program. Eight months later, the company resurrected the drug saying the analysis was “incorrect,” arguing that it was based on a smaller dataset that featured fewer patients who received high-dose aducanumab. Adding in the additional data showed aducanumab reduced clinical decline, the company said.

Biogen believes that, if approved, aducanumab would become the first treatment to slow decline in people with Alzheimer’s disease, a neurodegenerative disorder whose treatments focus on controlling symptoms. The data, however, are not so clear. In Yee’s words, “the phase 3 data was largely mixed and still inconclusive.”

RELATED: In a test for Hahn’s FDA, Biogen submits controversial Alzheimer’s drug aducanumab

The data underpinning the filing come from a phase 1b study, as well as a pair of phase 3 studies that tested aducanumab in patients with early-stage and mild Alzheimer’s. The phase 3 results were mixed, with one study suggesting that aducanumab is no better than placebo, while the other linked the drug to improved scores on a dementia scale.

Patients who received the highest dose of aducanumab in a phase 3 trial dubbed EMERGE showed statistically significant improvement on a clinical dementia scale. But the same patient group in the ENGAGE phase 3 study did worse than patients taking placebo on that same measure, as well as on a test of cognitive function.

RELATED: 2019’s top 15 clinical trial flops (and a flip-flop) | Flip-flop: Aducanumab

Biogen did not use a priority review voucher to snag a speedy review, “suggesting FDA sees this as an unmet need and willing to review this under an accelerated window,” Yee wrote in an investor note.

Optimists argue that the speedy review suggests the agency is “comfortable with the totality of the data, recognizes high unmet need, and really wants to get an Alzheimer’s drug approved,” Yee wrote. Others might read less into the priority review, as the FDA could still ask Biogen for more data or extend their review period.

Some analysts have been cautiously optimistic. Cantor Fitzgerald’s Alethia Young wrote earlier this year that she was “reasonably confident” the FDA would accept aducanumab’s application, adding the caveat that she and her colleagues “don’t view this approval as certain,” given “the complicated issues around the statistical analysis and the inconsistent outcomes from both pivotal trials.” Brian Abrahams of RBC Capital Markets estimated a 30% chance the FDA would approve the drug “despite the mixed ph.III data.”

But Baird analyst Brian Skorney has dismissed aducanumab’s chances, saying the bulk of the data shows aducanumab doesn’t provide a clinical benefit.

“If Biogen said based on this data it was running another study, the stock would be down because the data doesn’t justify investing in another study. The regulatory threshold is higher than that,” Skorney wrote in December. “The bottom line is, the FDA standard of approval is substantial evidence of efficacy and the cumulative data for aducanumab falls really far of this standard.”


Continue Reading
AI4 hours ago

Impacts of Artificial Intelligence in Content Writing

Publications6 hours ago

Hong Kong media tycoon Jimmy Lai arrested under security law

Cannabis6 hours ago

How to make a cannabis-infused canna-grapefruit spritz

Cannabis6 hours ago

Could Joe Biden budge on cannabis legalization?

Cannabis6 hours ago

4 weed products Weldon Angelos can’t live without

Blockchain7 hours ago

A $15K Bitcoin Likely As Price Breaks Above “Multi-Year Bullish Triangle”

Blockchain7 hours ago

Ethereum Classic Under Multiple 51% Attacks | Bitcoin News Summary Aug 10, 2020

Publications7 hours ago

Chinese Tesla rival Xpeng Motors files for New York IPO

Publications8 hours ago

U.S. health chief offers Taiwan ‘strong’ support in landmark visit

Publications8 hours ago

Stock futures mixed after Trump signs orders extending coronavirus relief

Blockchain8 hours ago

Number of Bitcoin Cash Whales Drop Following 39% Price Surge

Blockchain8 hours ago

Bitcoin Price Tackles $12,000 After Breaking Through a Key Resistance Zone

Blockchain9 hours ago

Japanese Messaging Giant LINE’s LN Token Trading on BitMax

Blockchain10 hours ago

Bitcoin Erupts Past $12,000: Here’s What Analysts Think Comes Next

Publications10 hours ago

Some office space could get permanently cut during the pandemic. Here’s how companies will cope

Blockchain11 hours ago

Here’s Why Analysts Are Expecting For Ethereum To Drop Back Towards $370

Publications12 hours ago

Pelosi slams Trump’s executive actions on coronavirus relief: ‘Absurdly unconstitutional’

Automotive13 hours ago

Hyundai launches Ioniq as a standalone brand to exclusively make electric cars

Blockchain13 hours ago

How Miners Can Hedge Their Inventory to Increase Return on Investment

Blockchain14 hours ago

Analysts Expect Chainlink (LINK) Reversal After 50% Eruption to $14

Blockchain14 hours ago

BAND Token is Now Available for Trading on Huobi Global

Publications15 hours ago

Amazon reportedly discussing using former J.C. Penney and Sears stores as fulfillment centers

Blockchain15 hours ago

DeFi has more than just yield farming to thank for the recent surge

Automotive15 hours ago

Judge denies bail for men accused of sneaking Carlos Ghosn out of Japan

Blockchain16 hours ago

What Hope Do Bears Have If Bitcoin Holds $11,500? Analyst Asks

Blockchain17 hours ago

Cardano short/medium-term price analysis: August 09

Publications17 hours ago

U.S. tops 5 million coronavirus cases as outbreak threatens America’s Midwest

Automotive18 hours ago

School buses are another coronavirus question mark

Blockchain18 hours ago

Will Bitcoin be the go-to asset during the incoming stagflation?

Cannabis18 hours ago

5 Thing You Can Do To Make Your Weeks Run Smoother

Automotive19 hours ago

Max Verstappen wins 70th Anniversary Grand Prix at Silverstone

Blockchain19 hours ago

Economic Crisis Leaves US Government Officials in State of Confusion

Blockchain19 hours ago

Litecoin short-term price analysis: 09 August

Automotive19 hours ago

This ‘Hoverboard’ can transform into a rideable 4-wheeler

Blockchain20 hours ago

Bitcoin: What to expect during institutional ‘land grab’ phase?

Blockchain20 hours ago

LINK Trading Volume Surpasses Bitcoin on Coinbase

Publications21 hours ago

While some techies flee Silicon Valley, this Waymo engineer is doubling down and running for office

Blockchain21 hours ago

Bitcoin’s price surge has depleted long-term hodlings

Blockchain22 hours ago

The Top Dice Strategies That Actually Make You Money

Cannabis22 hours ago

Stanley Brothers Face Another Setback with Final Refusal of “CW” Trademark