June 2018 - Lisa_seminaires - mailman.iro.umontreal.ca

[Tea Talk] Anne Churchland (Cold Spring Harbor Laboratory) June 29 2018 10:30 AA3195
by Michael Noukhovitch 29 Jun '18

29 Jun '18

This week we have *Anne Churchland * from * Cold Spring Harbor Laboratory * giving a talk on *June 29 2018* at *10:30* in room *AA3195* This talk may be streamed as usual here <https://mila.bluejeans.com/1365483656/webrtc> And you can sign up to meet the speaker here <https://calendar.google.com/calendar/selfsched?sstoken=UVBVTVF0X25Nd09LfGRl…> Don't even think about it! Decide to come to the tea talk and let your feet spontaneously guide the way. Michael *TITLE* Spontaneous movements dominate cortical activity during sensory-guided decision making *KEYWORDS *neuroscience, decision-making, neuro-imaging *ABSTRACT* Animals continually produce a wide array of spontaneous and learned movements and undergo rapid internal state transitions. Most work in neuroscience ignores this “internal backdrop” and instead focuses on neural activity aligned to task-imposed variables such as sensory stimuli. We sought to understand the joint effects of internal backdrop vs. task-imposed variables. We measured neural activity using calcium imaging via a widefield macroscope during decision-making. Surprisingly, the impact of the internal backdrop dwarfed task-imposed sensory and cognitive signals. This dominance was comparable in novice and expert decision-makers and was even stronger in single neuron measurements from frontal cortex. These results highlight spontaneous and learned movements as the main determinant of large-scale cortical activity. By leveraging a wide array of animal movements, our model offers a very general method for separating the impact of internal backdrop from task-imposed neural activity. *BIO* Anne K. Churchland received her Ph.D. in neuroscience from the University of California, San Francisco, advised by Dr. Stephen Lisberger. She then did a postdoctoral fellowship with Dr. Michael Shadlen at the University of Washington in the Physiology and Biophysics Department. Her postdoctoral work focused on mechanisms of decision making in nonhuman primates and included both experimental and theoretical work. The latter was funded by a Pathways to Independence (K99) Award from the National Eye Institute. In 2010, she became an assistant professor at Cold Spring Harbor Laboratory. In starting her own laboratory, Professor Churchland began studying decision making using rodent models to take advantage of emerging tools for circuit dissection which are readily available in rodents. Since then, her laboratory has been a major player in bringing behavioral paradigms to rodents that have been successful in elucidating neural mechanisms in primates. These include perceptual decision making and multisensory integration. Most recently, she has teamed up with other 20 other researchers to for the International Brain Laboratory, a collaboration between theorists and experimentalists aiming to understand the neural computations supporting decision-making. Since joining Cold Spring Harbor Laboratory, Professor Churchland has been the recipient of awards from the McKnight Foundation, the Pew Charitable Trusts, the Klingenstein-Simons Foundation, the Simons Collaboration on the Global Brain, the John Merck Fund and the Chapman Foundations. In addition to her scientific work, Professor Churchland runs science outreach activities at public schools and co-directed the Undergraduate Research Program at Cold Spring Harbor Laboratory. She also maintains a blog about neuroscience research that is directed at a mainstream audience; its recent focus is on reviewing papers on the preprint server. Professor Churchland also funded anneslist.net, a resource highlighting women in systems and computational neuroscience. She received the 2017 Louise Hanson Marshall Special Recognition Award from the Society for Neuroscience for promoting diversity in the field and the 2012 Janett Rosenberg Trubatch Career Development Award for creativity in research prior to tenure.

2 4

[Tea Talk] Dzmitry Bahadanau (MILA) Fri June 22 10:30AM AA3195
by Michael Noukhovitch 22 Jun '18

22 Jun '18

This week we have our very own *Dzmitry Bahadanau* giving a talk on work he did on internship at DeepMind on *Friday June 22* at *10:30AM* in room *AA3195*. Help Dima get a human baseline for instruction following and *come to the tea talk* Michael *TITLE* Learning to Follow Language Instructions with Adversarial Reward Induction *KEYWORDS *language understanding, reinforcement learning, grounded language *ABSTRACT* Recent work has shown that deep reinforcement-learning agents can learn to follow language-like instructions from infrequent environment rewards. However, for many real-world natural language commands that involve a degree of underspecification or ambiguity, such as *tidy the room*, it would be challenging or impossible to program an appropriate reward function. To overcome this, we present a method for learning to follow commands from a training set of instructions and corresponding example goal-states, rather than an explicit reward function. Importantly, the example goal-states are not seen at test time. The approach effectively separates the representation of what instructions require from how they can be executed. In a simple grid world, the method enables an agent to learn a range of commands requiring interaction with blocks and understanding of spatial relations and underspecified abstract arrangements. We further show the method allows our agent to adapt to unseen instructions and changes in the environment without requiring new training examples. *BIO* Dzmitry Bahdanau is a Ph.D. student at the Montreal Institute for Learning Algorithms under the supervision of Yoshua Bengio. His research interest is enabling intelligent assistants that could collaborate with humans and communicate with them in natural language. In the view of this long-term research goal he is interested in language acquisition, language grounding models, reinforcement learning and imitation learning. Dzmitry studied applied mathematics at Belarusian State University in Minsk and obtained his MsC degree in computer science at Jacobs University Bremen, Germany.

1 1

[Tea Talk] Yaroslav Ganin (MILA) Fri June 15 10:30AM AA3195
by Michael Noukhovitch 16 Jun '18

16 Jun '18

This week we have our very own Yaroslav Ganin giving a talk on the work he did on internship at DeepMind on Friday June 15th at 10:30AM in room AA3195 This talk may be streamed as usual here <https://bluejeans.com/809027115/webrtc> And you can sign up to meet the speaker by messaging him on slack If you don't come to the talk, be prepared for a strongly reinforced adversary in the form of me! Michael *TITLE* Synthesizing Programs for Images using Reinforced Adversarial Learning *KEYWORDS *GAN, Reinforcement Learning, Unsupervised Learning *ABSTRACT * Advances in deep generative networks have led to impressive results in recent years. Nevertheless, such models can often waste their capacity on the minutiae of datasets, presumably due to weak inductive biases in their decoders. This is where graphics engines may come in handy since they abstract away low-level details and represent images as high-level programs. Current methods that combine deep learning and renderers are limited by hand-crafted likelihood or distance functions, a need for large amounts of supervision, or difficulties in scaling their inference algorithms to richer datasets. To mitigate these issues, we present SPIRAL, an adversarially trained agent that generates a program which is executed by a graphics engine to interpret and sample images. The goal of this agent is to fool a discriminator network that distinguishes between real and rendered data, trained with a distributed reinforcement learning setup without any supervision. A surprising finding is that using the discriminator’s output as a reward signal is the key to allow the agent to make meaningful progress at matching the desired output rendering. To the best of our knowledge, this is the first demonstration of an end-to-end, unsupervised and adversarial inverse graphics agent on challenging real world (MNIST, Omniglot, CelebA) and synthetic 3D datasets

2 3

[Tea Talk] Maksym Korablyov (MIT) Thr June 7 2018 2:00PM AA3195
by Michael Noukhovitch 07 Jun '18

07 Jun '18

Sorry for the delay but this week we have *Maksym Korablyov * from * MIT * giving a talk on *Thursday June 7 2018* at *2:00PM* in room *AA3195. *This is usually the learning agents' reading group slot so shoutout to them for letting me have this slot and again to Dima for helping put this together. Maksym is interested in meeting with people to discuss RNNs for structural prediction, conditional feature extraction, and 3D object alignment, and also other applications of deep learning to chemistry. Want to meet with him? Sign up here <https://calendar.google.com/calendar/selfsched?sstoken=UVBVTVF0X25Nd09LfGRl…> Can't make it? Email me and I'll set something up! Note: There will be no streaming so come in person! If your *deep learning* code is *breaking bad*, take a break and come to this talk! Michael *TITLE* A few steps towards molecule google *KEYWORDS *deep learning chemistry, deep learning theory *ABSTRACT* Commonly used search engines such as Google can retrieve relevant entries with high speed and accuracy from spaces as big as 10^12 - the number of words in the Internet. Search algorithms that could address the space of 10^80 - 10^100 drug-like molecules are not yet publicly available. The world’s combined effort of pharmaceutical companies (experimental + computational) is ~10^15 molecules/year with 30-50 new drugs found. In this talk I will 1: give some numerical estimates of the value of unexplored chemical space. 2: provide some intuition why all possible molecules could be mapped to a low-dimensional manifold and which families of functions might efficiently describe such mapping. 3: show some empirical results of the search in 10^80. 4: present an algorithm for partial flexible shape match (drug + protein) that we have developed recently. *BIO* Maksym Korablyov is a graduate student advised by Dr. Joseph Jacobson at Massachusetts Institute of Technology, Media Lab. Prior to MIT, Maksym was a Fulbright Fellow at Harvard Medical School with Dr. Gil Alterovitz for one year. Maksym holds MS in Bioinformatics from Georgia Institute of Technology where he developed a series of potent antiviral inhibitors jointly with Dr. James Gumbart and Dr. Raymond Schinazi.

2 1