Understanding reality through algorithms | MIT News

Although Fernanda De La Torre has a few years left in her graduate studies, she is already dreaming big for her future.

“I have a dream of one day starting a school where I can take this world of understanding and perception to a place where I would never have access to it,” she said.

It’s this kind of ambitious thinking that got De La Torre, a doctoral student in MIT’s Department of Brain and Cognitive Sciences, so far. De La Torre, who recently received the prestigious Paul and Daisy Soros New American Fellowship, discovered a supportive, creative research environment at MIT that allowed her to delve into the cutting-edge science of artificial intelligence. But she remains driven by an innate curiosity about the human imagination and a desire to bring that knowledge to the community in which she grew up.

An unconventional path to neuroscience

De La Torre’s first exposure to neuroscience was not in the classroom, but in her daily life. As a child, she watched her sister struggle with epilepsy. At age 12, she illegally entered the United States from Mexico to reunite with her mother, exposing her to a new language and culture. Once in America, she had to deal with her mother’s changing personality in an abusive relationship. “All these different things I see around me drive me to want to better understand how psychology works,” de la Torre said, “to understand how the mind works and how we can all be in the same The environment and feel are very different.”

But finding an outlet for this intellectual curiosity is challenging. As an undocumented immigrant, she had limited access to financial aid. Her high school was also underfunded and lacked electives. However, mentors along the way encouraged the aspiring scientist, and through a program at her school, she was able to enroll in community college classes to meet basic educational requirements.

There was an inspiring dedication to her education, but De La Torre managed to get into Kansas State University as an undergraduate, majoring in computer science and mathematics. At Kansas State University, she had her first real experience with research. “I was just fascinated by the questions they asked and the whole space I hadn’t encountered,” De La Torre said of her experience working in the Visual Cognition Lab and discovering the field of computational neuroscience.

Although Kansas State does not have a dedicated neuroscience program, her research experience in cognition led her to a machine learning lab led by computer science professor William Hsu. There, De La Torre became fascinated by the possibility of using computing to simulate the human brain. Xu’s support also convinced her that a career in science was possible. “He always made me feel like I was capable of solving big problems,” she said fondly.

Building on the confidence she had at Kansas State University, De La Torre came to MIT in 2019 as a post-baccalaureate student at Tomaso Poggio, Eugene McDermott Professor of Brain and Cognitive Sciences and McGovern Institute for Brain Research Fellow Tomaso Poggio Lab. Along with Poggio, who is also director of the Center for Brains, Minds, and Machines, De La Torre began work on deep learning theory, a field of machine learning focused on how artificial neural networks modeled on the brain learn to recognize patterns and learn .

“It’s a really interesting problem because we’re starting to use them everywhere,” says De La Torre of Neural Networks, citing examples from self-driving cars to medicine. “But, at the same time, we don’t fully understand how these networks go from knowing nothing, just a bunch of numbers, to outputting something meaningful.”

Her experience as a postgraduate was De La Torre’s first real opportunity to apply the computer technology skills she developed as an undergraduate to neuroscience. It was also the first time she could fully focus on research. “That was the first time I had health insurance and a steady paycheck. It was a life-changing approach in itself,” she said. “But in terms of research, it was very daunting at first. I was anxious and I wasn’t sure if I belonged here.”

Fortunately, De La Torre says she was able to overcome these insecurities, both thanks to a growing unabashed enthusiasm for the field, and thanks to Poggio and her faculty at MIT’s Department of Brain and Cognitive Sciences. Support from other colleagues. When given the opportunity to apply for the department’s doctoral program, she embraced it. “Just knowing that these mentors are here and they care about their students,” De La Torre said of her decision to stay at MIT for graduate school. “That really makes sense.”

Expanding the concept of reality and imagination

So far, during her two years in graduate school, De La Torre’s work has expanded her understanding of neural networks and their applications to the study of the human brain. In collaboration with Guangyu Yang, an associate research fellow at the McGovern Institute and an assistant professor in the Departments of Brain and Cognitive Sciences, Electrical Engineering, and Computer Science, she is working on what she describes as the more philosophical questions about how a person develops self-awareness as an independent The presence. She was intrigued by how this sense of self developed and why it might be useful.

However, De La Torre’s main advisor is Professor Josh McDermott, who runs the Computational Audition Lab. With McDermott, De La Torre is trying to understand how the brain integrates sight and sound. While combining sensory input appears to be a fundamental process, there are many unanswered questions about how our brains combine multiple signals into a coherent impression or perception of the world. Many questions are raised by audiovisual illusions, in which what we hear changes what we see. For example, if a person sees a video of two discs passing each other, but the clip contains the sound of a collision, the brain perceives that the discs are bouncing off, rather than passing each other. Given an ambiguous image, this simple auditory cue is all it takes to create a different perception of reality.

“Something interesting is happening, our brains are picking up two signals that tell us different things, and yet, we have to somehow combine them to understand the world,” she said.

De La Torre is using behavioral experiments to explore how the human brain understands multisensory cues to construct specific perceptions. To do this, she created scenes in which various objects interacted with different sounds in 3D space, and asked study participants to describe the characteristics of the scene. In one experiment, for example, she combined the visual effect of blocks moving across a surface at different speeds with various scratching sounds, asking participants to estimate the roughness of the surface. Ultimately, she hopes to bring the experiment to virtual reality, where participants will physically push blocks in response to how rough they think the surface is, rather than just reporting their experiences.

Once she’s collected the data, she’ll move on to the modeling phase of the study to assess whether the multisensory neural network perceives illusions like humans do. “What we want to do is simulate exactly what’s going on,” De La Torre said. “How do we take these two signals, integrate them, and use all our prior knowledge and inferences from physics to truly understand the world?”

Although her two studies with Yang and McDermott appear to be distinct, she sees a clear connection between the two. Both projects are about mastering the capabilities of artificial neural networks and what they tell us about the brain. On a more fundamental level, she says how the brain perceives the world from different sensory cues may be part of what gives people self-awareness. Sensory perception is about building a cohesive, unified sense of the world from multiple sources of sensory data. Likewise, she argues, “Self-consciousness is really a combination of actions, plans, goals, emotions, all of which are parts of their own, but somehow create a single being.”

It’s an appropriate sentiment for De La Torre, who has been struggling to understand and integrate different aspects of his life. For example, while working at the Computational Audition lab, she began experimenting with combining electronic music with folk music from her native Mexico, bridging her “two worlds,” as she puts it. Having the space for this kind of intellectual exploration, and colleagues who encourage it, is one of De La Torre’s favorite parts of MIT.

“I’m amazed at how many students think beyond professors,” she said. “I see a lot of goodness and excitement about science, and one more thing — it’s not a nerd, it’s a love of niche things — I just kind of like it.”

Source link