Current transfer learning methods are mainly based on fine tuning a pretrained model with target-domain data. Motivated by the techniques from adversarial machine learning (ML) that are capable of manipulating the model prediction via data perturbations, in this paper we propose a novel approach, black-box adversarial reprogramming (BAR), that repurposes a well-trained black-box ML model (e.g., a prediction API or a proprietary software) for solving different ML tasks,especially in the scenario with scarce data and constrained resources. The rationale lies in exploiting high-performance but unknown ML models to gain learning capability for transfer learning.Using zeroth order optimization and multi-label mapping techniques, BAR can reprogram a black-box ML model solely based on its input-output responses without knowing the model architecture or changing any parameter. More importantly, in the limited medical data setting, on autism spectrum disorder classification, diabetic retinopathy detection, and melanoma detection tasks, BAR outperforms state-of-the-art methods and yields comparable performance to the vanilla adversarial reprogramming method requiring complete knowledge of the target ML model. BAR also out-performs baseline transfer learning approaches by a significant margin, demonstrating cost-effective means and new insights for transfer learning. Read More
Monthly Archives: August 2020
What happens in Vegas… is captured on camera
Police departments around the US use face recognition in highly varied ways, with little consensus on best practices.
The use of facial recognition by police has come under a lot of scrutiny. In part three of our four-part series on FaceID, host Jennifer Strong takes you to Sin City, which actually has one of America’s most buttoned-up policies on when cops can capture your likeness. She also finds out why celebrities like Woody Harrelson are playing a starring role in conversations about this technology. Read More
Too many AI researchers think real-world problems are not relevant
Any researcher who’s focused on applying machine learning to real-world problems has likely received a response like this one: “The authors present a solution for an original and highly motivating problem, but it is an application and the significance seems limited for the machine-learning community.”
These words are straight from a review I received for a paper I submitted to the NeurIPS (Neural Information Processing Systems) conference, a top venue for machine-learning research. I’ve seen the refrain time and again in reviews of papers where my coauthors and I presented a method motivated by an application, and I’ve heard similar stories from countless others. Read More
China Won’t Win the Race for AI Dominance
Once upon a time, Japan was widely expected to eclipse the United States as the technological leader of the world. In 1988, the New York Times reporter David Sanger described a group of U.S. computer science experts, meeting to discuss Japan’s technological progress. When the group assessed the new generation of computers coming out of Japan, Sanger wrote, “any illusions that America had maintained its wide lead evaporated.”
Replace “computers” with “artificial intelligence,” and “Japan” with “China,” and the article could have been written today. Read More
Mainframes: A Provisional Analysis of Rhetorical Frames in AI
When it comes to artificial intelligence, the headlines suggest that great powers are engaged in an AI arms race: “For Superpowers, Artificial Intelligence Fuels New Global Arms Race,” reads one story in Wired. “China is Winning a New Global Arms Race,” observes Bloomberg Markets and Finance. One report in The Wall Street Journal asserts, “The New Arms Race in AI.” To what degree do these headlines accurately represent elite opinion about AI?
Framing technological competition as an “AI arms race” or “battle for supremacy” has implications for policy, security, and international cooperation. Read More
A Radical New Model of the Brain Illuminates Its Wiring
Network neuroscience could revolutionize how we understand the brain—and change our approach to neurological and psychiatric disorders.
In mid-19th century Europe, a debate was raging among early brain scientists. Strangely, this academic disagreement had its roots in the pseudoscience of phrenology, the practice of measuring bumps on the skull to determine someone’s personality. Phrenology had found purchase at fairs and was quite popular with the general public, but it had been roundly rejected by most scholars. For others, though, this carnival trick held a pearl of inspiration. Phrenology depended on the assumption that different parts of the brain are associated with different traits and abilities, a position called “localizationism.” And the absurdity of skull-measuring did not necessarily invalidate this notion.
But others disliked the stench of charlatanism that clung to any ideas associated with phrenology. This second camp contended that capacities are evenly distributed throughout the brain, and so damage to any one brain region would have the same effect as damage to any other. The debate between these groups raged until 1861, when Paul Broca, a French neurologist, reported on a patient with a bizarre set of symptoms. Though this man could not speak, he was entirely capable of understanding language, and his intelligence seemed unaffected. When the patient died and Broca dissected his brain, he discovered a lesion, or site of severe damage, low on the left side of his brain. Here was an individual who had sustained brain damage in a specific area and had lost a very specific ability—while the rest of his functions remained intact! Localizationism had been vindicated. For the next 150 years, it would be the dominant position in brain science. Read More
There is a crisis of face recognition and policing in the US
When news broke that a mistaken match from a face recognition system had led Detroit police to arrest Robert Williams for a crime he didn’t commit, it was late June, and the country was already in upheaval over the death of George Floyd a month earlier. Soon after, it emerged that yet another Black man, Michael Oliver, was arrested under similar circumstances as Williams. While much of the US continues to cry out for racial justice, a quieter conversation is taking shape about face recognition technology and the police. We would do well to listen.
When Jennifer Strong and I started reporting on the use of face recognition technology by police for our new podcast, “In Machines We Trust,” we knew these AI-powered systems were being adopted by cops all over the US and in other countries. But we had no idea how much was going on out of the public eye. Read More
A Berkeley computer science student used GPT-3 to generate a No. 1 blog entry
Liam Porr collaborated with a Berkeley Ph.D. student to use GPT-3 to generate completed posts for the Adolos blog. A post titled “Feeling unproductive? Maybe you should stop overthinking” reached No. 1 on Hacker News. Porr retired the blog after two weeks with a final, cryptic post titled “What I would do with GPT-3 if I had no ethics” Read More
#fake, #nlpNew AI Dupes Humans into Believing Synthesized Sound Effects Are Real
Using machine-learning, AutoFoley determines what actions are taking place in a video clip and creates realistic sound effects.
… Researchers have created an automated program that analyzes the movement in video frames and creates its own artificial sound effects to match the scene. In a survey, the majority of people polled indicated that they believed the fake sound effects were real. The model, AutoFoley, is described in a study published June 25 in IEEE Transactions on Multimedia. Read More
Build a Hamilton Song Recommendation SMS Bot with Machine Learning
Hamilton the Musical will start streaming on Disney Plus this Friday, so happy Hamilfilm week! To celebrate, learn how to build a SMS chatbot that recommends the Hamilton song most relevant to you right now using Twilio Programmable SMS and Functions, Microsoft Azure Cognitive Services, and JavaScript.
See it in-action: text how you’re feeling to +13364295064. The longer and more descriptive your message is, the more data the app has to analyze what Hamilton song you need now! Read Now
Rick's Cafe AI 