BEGIN:VCALENDAR
VERSION:2.0
PRODID:researchseminars.org
CALSCALE:GREGORIAN
X-WR-CALNAME:researchseminars.org
BEGIN:VEVENT
SUMMARY:Jason Weston (Facebook)
DTSTART:20201005T190000Z
DTEND:20201005T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/1
DESCRIPTION:Title: LIGHT: Training Agents that can Act and Speak with Other M
odels and Humans in a Rich Text Adventure Game World\nby Jason Weston
(Facebook) as part of Machine Learning Advances and Applications Seminar\n
\n\nAbstract\nThe title just about covered it\, but.. LIGHT is a rich fant
asy text adventure game environment featuring dialogue and actions between
agents in the world\, which consist of both models and humans.\n\nI will
summarize work on this platform\, including crowdsourcing and machine lear
ning to build the rich world environment\, training agents to speak and ac
t within it\, and deploying the game for lifelong learning of agents by in
teracting with humans.\n\nThis is joint work with a number of authors: Emi
ly Dinan\, Angela Fan\, Samuel Humeau\, Saachi Jain\, Siddharth Karamcheti
\, Douwe Kiela\, Margaret Li\, Shrimai Prabhumoye\, Emma Qian\, Tim Rockt
äschel\, Pratik Ringshia\, Kurt Shuster\, Arthur Szlam\, Adina Williams.
See https://parl.ai/projects/light/ for papers & more info!\n\nBio: Jason
Weston is a Research Scientist at Facebook\, NY and a Visiting Research Pr
ofessor at NYU. He earned his PhD in Machine Learning at Royal Holloway\,
University of London and at AT&T Research in Red Bank\, NJ (advisors: Alex
Gammerman\, Volodya Vovk and Vladimir Vapnik) in 2000. From 2000 to 2001\
, he was a Researcher at Biowulf Technologies. From 2002 to 2003 he was a
Research Scientist at the Max Planck Institute for Biological Cybernetics
in Tuebingen\, Germany. From 2003 to 2009 he was a research staff member a
t NEC Labs America\, Princeton. From 2009 to 2014 he was a Research Scient
ist at Google\, NY. His interests lie in statistical machine learning\, wi
th a focus on reasoning\, memory\, perception\, interaction and communicat
ion. Jason has published over 100 papers\, including best paper awards at
ICML and ECML\, and a Test of Time Award for his work "A Unified Architect
ure for Natural Language Processing: Deep Neural Networks with Multitask L
earning\," ICML 2008 (with Ronan Collobert). He was part of the YouTube te
am that won a National Academy of Television Arts & Sciences Emmy Award fo
r Technology and Engineering for Personalized Recommendation Engines for V
ideo Discovery. He was listed as the 16th most influential machine learnin
g scholar at AMiner and one of the top 50 authors in Computer Science in S
cience.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/1/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Sham Kakade (University of Washington)
DTSTART:20201102T200000Z
DTEND:20201102T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/2
DESCRIPTION:Title: Policy Gradient Methods\, Curvature\, and Distribution Shi
ft\nby Sham Kakade (University of Washington) as part of Machine Learn
ing Advances and Applications Seminar\n\n\nAbstract\nReinforcement learnin
g is now the dominant paradigm for how an agent learns to interact with th
e world in order to achieve some long term objectives. Here\, policy gradi
ent methods are among the most effective methods in challenging reinforcem
ent learning problems\, due to that they: are applicable to any differenti
able policy parameterization\; admit easy extensions to function approxima
tion\; easily incorporate structured state and action spaces\; are easy to
implement in a simulation based\, model-free manner.\n\nHowever\, little
is known about even their most basic theoretical convergence properties\,
including:\n\n- do they converge to a globally optimal solution\, say with
a sufficiently rich policy class?\n\n- how well do they cope with approxi
mation error\, say due to using a class of neural policies?\n\n- what is t
heir finite sample complexity?\n\nThis talk will survey a number of result
s on these basic questions. We will highlight the interplay of theory\, al
gorithm design\, and practice.\n\nJoint work with: Alekh Agarwal\, Jason L
ee\, Gaurav Mahajan\n\nBio: Sham Kakade is a professor in the Department o
f Computer Science and the Department of Statistics at the University of W
ashington and is also a senior principal researcher at Microsoft Research.
His work is on the mathematical foundations of machine learning and AI. S
ham's thesis helped lay the statistical foundations of reinforcement learn
ing. With his collaborators\, his additional contributions include: one of
the first provably efficient policy search methods in reinforcement learn
ing\; developing the mathematical foundations for the widely used linear b
andit models and the Gaussian process bandit models\; the tensor and spect
ral methodologies for provable estimation of latent variable models\; the
first sharp analysis of the perturbed gradient descent algorithm\, along w
ith the design and analysis of numerous other convex and non-convex algori
thms. He is the recipient of the ICML Test of Time Award\, the IBM Pat Gol
dberg best paper award\, and INFORMS Revenue Management and Pricing Prize.
He has been program chair for COLT 2011.\n\nSham was an undergraduate at
Caltech\, where he studied physics and worked under the guidance of John P
reskill in quantum computing. He completed his Ph.D. with Peter Dayan in c
omputational neuroscience at the Gatsby Unit at University College London.
He was a postdoc with Michael Kearns at the University of Pennsylvania.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/2/
END:VEVENT
BEGIN:VEVENT
SUMMARY:David Lopez-Paz (Facebook)
DTSTART:20201130T200000Z
DTEND:20201130T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/3
DESCRIPTION:by David Lopez-Paz (Facebook) as part of Machine Learning Adva
nces and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/3/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Douglas Eck (Google)
DTSTART:20201214T200000Z
DTEND:20201214T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/4
DESCRIPTION:Title: Challenges in Building ML Algorithms for the Creative Comm
unity\nby Douglas Eck (Google) as part of Machine Learning Advances an
d Applications Seminar\n\n\nAbstract\nMagenta is an open-source project ex
ploring the role of machine learning as a tool in the creative process. We
've been running in public (g.co/magenta) for over four years. This talk w
ill look back at successes and frustrations in bringing our work to creato
rs\, mostly musicians. I'll also talk about some current and future work.
Magenta is made up of several ML researchers and engineers on the Google B
rain team\, which focuses on deep learning. Our successes have mostly been
in the area of new algorithm development (NSynth\, MusicVAE\, Music Trans
former\, DDSP and others). Our frustrations have been in finding ways to m
ake these models useful for music creators. The talk will be a casual exam
ple-driven discussion about what worked and what didn't\, and where we're
going next. Spoiler: we have been humbled by the user interface challenges
encountered when building tools for creative work. My main message for Ve
ctor Institute is that machine learning alone is not enough to address a c
hallenge like enabling new forms of creativity -- you need to think about
what artists really want and how to communicate with them.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/4/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Moritz Hardt (UC Berkeley)
DTSTART:20210125T200000Z
DTEND:20210125T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/6
DESCRIPTION:Title: Performative Prediction\nby Moritz Hardt (UC Berkeley)
as part of Machine Learning Advances and Applications Seminar\n\n\nAbstra
ct\nWhen predictive models support decisions they can influence the outcom
e they aim to predict. We call such predictions performative\; the predict
ion influences the target. Performativity is a well-studied phenomenon in
policy-making that has so far been neglected in supervised learning. When
ignored\, performativity surfaces as undesirable distribution shift\, rout
inely addressed with retraining. \n\nIn this talk\, I will describe a risk
minimization framework for performative prediction bringing together conc
epts from statistics\, game theory\, and causality. A new element is an eq
uilibrium notion called performative stability. Performative stability imp
lies that the predictions are calibrated not against past outcomes\, but a
gainst the future outcomes that manifest from acting on the prediction. \n
\nI will then discuss recent results on performative prediction including
necessary and sufficient conditions for the convergence of retraining to a
performatively stable point of nearly minimal loss. \n\nJoint work with J
uan C. Perdomo\, Tijana Zrnic\, and Celestine Mendler-Dünner.\n\nBio: Mor
itz Hardt is an Assistant Professor in the Department of Electrical Engine
ering and Computer Sciences at the University of California\, Berkeley. Ha
rdt investigates algorithms and machine learning with a focus on reliabili
ty\, validity\, and societal impact. After obtaining a PhD in Computer Sci
ence from Princeton University\, he held positions at IBM Research Almaden
\, Google Research and Google Brain. Hardt is a co-founder of the Workshop
on Fairness\, Accountability\, and Transparency in Machine Learning (FAT/
ML) and a co-author of the forthcoming textbook "Fairness and Machine Lear
ning". He has received an NSF CAREER award\, a Sloan fellowship\, and best
paper awards at ICML 2018 and ICLR 2017.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/6/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Peter Dayan (Max Planck Institute)
DTSTART:20210208T200000Z
DTEND:20210208T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/7
DESCRIPTION:Title: Peril\, Prudence and Planning as Risk\, Avoidance and Worr
y\nby Peter Dayan (Max Planck Institute) as part of Machine Learning A
dvances and Applications Seminar\n\n\nAbstract\nRisk occupies a central ro
le in both the theory and practice of decision-making. Although it is deep
ly implicated in many conditions involving dysfunctional behavior and thou
ght\, modern theoretical approaches to understanding and mitigating risk i
n either one-shot or sequential settings have yet to permeate fully the fi
elds of neural reinforcement learning and computational psychiatry. I will
discuss the use of one prominent approach\, called conditional value-at-r
isk to examine both the nature of risk avoidant choices\, encompassing suc
h things as justified gambler's fallacies\, and the optimal planning that
can lead to consideration of such choices\, with implications for offline\
, ruminative\, thinking in the context of anxiety.\n\nThis is joint work w
ith Chris Gagne.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/7/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Chelsea Finn (Stanford/Google)
DTSTART:20210222T200000Z
DTEND:20210222T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/8
DESCRIPTION:Title: Principles for Tackling Distribution Shift: Pessimism\, Ad
aptation\, and Anticipation\nby Chelsea Finn (Stanford/Google) as part
of Machine Learning Advances and Applications Seminar\n\n\nAbstract\nWhil
e we have seen substantial progress in machine learning\, a critical short
coming of current methods lies in handling distribution shift between trai
ning and deployment. Distribution shift is pervasive in real-world problem
s ranging from natural variation in the distribution over locations or dom
ains\, to shifts in the distribution arising from different decision makin
g policies\, to shifts over time as the world changes. In this talk\, I'll
discuss three general principles for tackling these forms of distribution
shift: pessimism\, adaptation\, and anticipation. I'll present the most g
eneral form of each principle before providing concrete instantiations of
using each in practice. This will include a simple method for substantiall
y improving robustness to spurious correlations\, a framework for quickly
adapting a model to a new user or domain with only unlabeled data\, and an
algorithm that enables robots to anticipate and adapt to shifts caused by
other agents.\n\nBio: Chelsea Finn is an Assistant Professor in Computer
Science and Electrical Engineering at Stanford University. Finn's research
interests lie in the capability of robots and other agents to develop bro
adly intelligent behavior through learning and interaction. To this end\,
her work has included deep learning algorithms for concurrently learning v
isual perception and control in robotic manipulation skills\, self-supervi
sed methods for learning a breadth of vision-based control tasks\, and met
a-learning algorithms that can enable fast\, few-shot adaptation in both v
isual perception and deep reinforcement learning. Finn received her Bachel
or's degree in Electrical Engineering and Computer Science at MIT and her
PhD in Computer Science at UC Berkeley. Her research has been recognized t
hrough the Microsoft Research Faculty Fellowship\, the ACM doctoral disser
tation award\, the C.V. Ramamoorthy Distinguished Research Award\, and the
MIT Technology Review 35 under 35 Award\, and her work has been covered b
y various media outlets\, including the New York Times\, Wired\, and Bloom
berg.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/8/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Percy Liang (Stanford Universi)
DTSTART:20210308T200000Z
DTEND:20210308T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/9
DESCRIPTION:Title: Deep Models and on Shaping their Development\nby Percy
Liang (Stanford Universi) as part of Machine Learning Advances and Applic
ations Seminar\n\n\nAbstract\nModels in deep learning are wild beasts: the
y devour raw data\, are powerful but hard to control. This talk explores t
wo approaches to taming them. First\, I will introduce concept bottleneck
networks\, in which a deep neural network makes a prediction via interpret
able\, high-level concepts. We show that such models can obtain comparable
accuracy with standard models\, while offering the unique ability for a h
uman to perform test-time interventions on the concepts. Second\, I will i
ntroduce prefix-tuning\, which allows one to harness the power of pre-trai
ned language models (e.g.\, GPT-2) for text generation tasks. The key idea
is to learn a continuous task-specific prefix that primes the language mo
del for the task at hand. Prefix-tuning obtains comparable accuracy to fin
e-tuning\, while only updating 0.1% of the parameters. Finally\, I will en
d with a broad question: what kind of datasets should the community develo
p to drive innovation in modeling approaches? Are size and realism necessa
ry attributes of a dataset? Could we have made all the modeling progress i
n NLP without SQuAD? As this counterfactual question is impossible to answ
er\, we perform a retrospective study on 20 modeling approaches and show t
hat even a small\, synthetic dataset can track the progress that was made
on SQuAD. While inconclusive\, this result encourages us to think more cri
tically about the value of datasets during their construction.\n\nBio: Per
cy Liang is an Associate Professor of Computer Science at Stanford Univers
ity (B.S. from MIT\, 2004\; Ph.D. from UC Berkeley\, 2011). His research s
pans many topics in machine learning and natural language processing\, inc
luding robustness\, interpretability\, semantics\, and reasoning. He is al
so a strong proponent of reproducibility through the creation of CodaLab W
orksheets. His awards include the Presidential Early Career Award for Scie
ntists and Engineers (2019)\, IJCAI Computers and Thought Award (2016)\, a
n NSF CAREER Award (2016)\, a Sloan Research Fellowship (2015)\, a Microso
ft Research Faculty Fellowship (2014)\, and multiple paper awards at ACL\,
EMNLP\, ICML\, and COLT.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/9/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Daphne Koller (Stanford University)
DTSTART:20210322T190000Z
DTEND:20210322T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/10
DESCRIPTION:Title: Machine Learning: A new approach to drug discovery\nb
y Daphne Koller (Stanford University) as part of Machine Learning Advances
and Applications Seminar\n\n\nAbstract\nModern medicine has given us effe
ctive tools to treat some of the most significant and burdensome diseases.
At the same time\, it is becoming consistently more challenging and more
expensive to develop new therapeutics. A key factor in this trend is that
the drug development process involves multiple steps\, each of which invol
ves a complex and protracted experiment that often fails. We believe that\
, for many of these phases\, it is possible to develop machine learning mo
dels to help predict the outcome of these experiments\, and that those mod
els\, while inevitably imperfect\, can outperform predictions based on tra
ditional heuristics. To achieve this goal\, we are bringing together high-
quality data from human cohorts\, while also developing cutting edge metho
ds in high throughput biology and chemistry that can produce massive amoun
ts of in vitro data relevant to human disease and therapeutic intervention
s. Those are then used to train machine learning models that make predicti
ons about novel targets\, coherent patient segments\, and the clinical eff
ect of molecules. Our ultimate goal is to develop a new approach to drug d
evelopment that uses high-quality data and ML models to design novel\, saf
e\, and effective therapies that help more people\, faster\, and at a lowe
r cost.\n\nBio: Daphne Koller is CEO and Founder of insitro\, a machine-le
arning enabled drug discovery company. Daphne is also co-founder of Engage
li\, was the Rajeev Motwani Professor of Computer Science at Stanford Univ
ersity\, where she served on the faculty for 18 years\, the co-CEO and Pre
sident of Coursera\, and the Chief Computing Officer of Calico\, an Alphab
et company in the healthcare space. She is the author of over 200 refereed
publications appearing in venues such as Science\, Cell\, and Nature Gene
tics. Daphne was recognized as one of TIME Magazine's 100 most influential
people in 2012. She received the MacArthur Foundation Fellowship in 2004
and the ACM Prize in Computing in 2008. She was inducted into the National
Academy of Engineering in 2011 and elected a fellow of the American Assoc
iation for Artificial Intelligence in 2004\, the American Academy of Arts
and Sciences in 2014\, and the International Society of Computational Biol
ogy in 2017.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/10/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Surya Ganguli (Stanford University)
DTSTART:20210405T190000Z
DTEND:20210405T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/11
DESCRIPTION:Title: Weaving together machine learning\, theoretical physics\,
and neuroscience\nby Surya Ganguli (Stanford University) as part of M
achine Learning Advances and Applications Seminar\n\n\nAbstract\nAn exciti
ng area of intellectual activity in this century may well revolve around a
synthesis of machine learning\, theoretical physics\, and neuroscience. T
he unification of these fields will likely enable us to exploit the power
of complex systems analysis\, developed in theoretical physics and applied
mathematics\, to elucidate the design principles governing neural systems
\, both biological and artificial\, and deploy these principles to develop
better algorithms in machine learning. We will give several vignettes in
this direction\, including: (1) determining the best optimization problem
to solve in order to perform regression in high dimensions\; (2) finding e
xact solutions to the dynamics of generalization error in deep linear netw
orks\; (3) derving the detailed structure of the primate retina by analyzi
ng optimal convolutional auto-encoders of natural movies\; (4) analyzing a
nd explaining the origins of hexagonal firing patterns in recurrent neural
networks trained to path-integrate\; (5) understanding the geometry and d
ynamics of high dimensional optimization in the classical limit of dissipa
tive many-body quantum optimizers.\n\nReferences:\n\nM. Advani and S. Gang
uli\, Statistical mechanics of optimal convex inference in high dimensions
\, Physical Review X\, 6\, 031034\, 2016.\n\nM. Advani and S. Ganguli\, An
equivalence between high dimensional Bayes optimal inference and M-estima
tion\, NeurIPS\, 2016.\n\nA.K. Lampinen and S. Ganguli\, An analytic theor
y of generalization dynamics and transfer learning in deep linear networks
\, International Conference on Learning Representations (ICLR)\, 2019.\n\n
H. Tanaka\, A. Nayebi\, N. Maheswaranathan\, L.M. McIntosh\, S. Baccus\, S
. Ganguli\, From deep learning to mechanistic understanding in neuroscienc
e: the structure of retinal prediction\, NeurIPS 2019.\n\nS. Deny\, J. Lin
dsey\, S. Ganguli\, S. Ocko\, The emergence of multiple retinal cell types
through efficient coding of natural movies\, Neural Information Processin
g Systems (NeurIPS) 2018.\n\nB. Sorscher\, G. Mel\, S. Ganguli\, S. Ocko\,
A unified theory for the origin of grid cells through the lens of pattern
formation\, NeurIPS 2019.\n\nY. Bahri\, J. Kadmon\, J. Pennington\, S. Sc
hoenholz\, J. Sohl-Dickstein\, and S. Ganguli\, Statistical mechanics of d
eep learning\, Annual Reviews of Condensed Matter Physics\, 2020.\n\nY. Ya
mamoto\, T. Leleu\, S. Ganguli and H. Mabuchi\, Coherent Ising Machines: q
uantum optics and neural network perspectives\, Applied Physics Letters 20
20.\n\nB.P. Marsh\, Y\, Guo\, R.M. Kroeze\, S. Gopalakrishnan\, S. Ganguli
\, J. Keeling\, B.L. Lev\n\nEnhancing associative memory recall and storag
e capacity using confocal cavity QED\, https://arxiv.org/abs/2009.01227.\n
\nBio: Surya Ganguli triple majored in physics\, mathematics\, and EECS at
MIT\, completed a PhD in string theory at Berkeley\, and a postdoc in the
oretical neuroscience at UCSF. He is now an associate professor of Applied
physics at Stanford where he leads the Neural Dynamics and Computation La
b. His research spans the fields of neuroscience\, machine learning and ph
ysics\, focusing on understanding and improving how both biological and ar
tificial neural networks learn striking emergent computations. He has been
awarded a Swartz-Fellowship in computational neuroscience\, a Burroughs-W
ellcome Career Award\, a Terman Award\, a NeurIPS Outstanding Paper Award\
, a Sloan fellowship\, a James S. McDonnell Foundation scholar award in hu
man cognition\, a McKnight Scholar award in Neuroscience\, a Simons Invest
igator Award in the mathematical modeling of living systems\, and an NSF c
areer award.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/11/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Jakob Foerster (University of Toronto and Vector Institute)
DTSTART:20210111T200000Z
DTEND:20210111T203000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/12
DESCRIPTION:Title: Zero-Shot (Human-AI) Coordination (in Hanabi) and Ridge R
ider\nby Jakob Foerster (University of Toronto and Vector Institute) a
s part of Machine Learning Advances and Applications Seminar\n\n\nAbstract
\nIn recent years we have seen fast progress on a number of zero-sum bench
mark problems in AI\, e.g. Go\, Poker and Dota. In contrast\, success in t
he real world requires humans to collaborate and communicate with others\,
in settings that are\, at least partially\, cooperative. Recently\, the c
ard game Hanabi has been established as a new benchmark environment to fil
l this gap. In particular\, Hanabi is interesting to humans since it is en
tirely focused on theory of mind\, i.e.\, the ability to reason over the i
ntentions\, beliefs and point of view of other agents when observing their
actions. This is particularly important in applications such as communica
tion\, assistive technologies and autonomous driving.\n\nWe start out by i
ntroducing the zero-shot coordination setting as a new frontier for multi-
agent research\, which is partially addressed by Other-Play\, a novel lear
ning algorithm which biases learning towards more human compatible policie
s.\n\nLastly we introduce Ridge Rider\, our brand new algorithm which addr
esses both zero-shot coordination and other optimization problems where th
e objective we care about can by definition not be evaluated during traini
ng time.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/12/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Shakir Mohamed (DeepMind)
DTSTART:20210419T190000Z
DTEND:20210419T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/13
DESCRIPTION:Title: AI and Shared Prosperity\nby Shakir Mohamed (DeepMind
) as part of Machine Learning Advances and Applications Seminar\n\n\nAbstr
act\nThe question we'll explore in this talk is how we can redirect the pa
th we are taking as AI designers towards the promotion of shared prosperit
y\, and in designing and deploying AI in ways where both the benefits\, an
d the risks\, of new technology are shared equally across society. I'll ex
plore these ideas by combining different elements of my recent work. I'll
start with a concrete problem of developing AI for detecting organ damage
in hospitals. This use-case will highlight the interconnected sociotechnic
al system that machine learning operates within. This system has sets of v
alues and politics that must contend with a colonial legacy and colonialit
y\, and I'll explore thinking on decolonial AI\, and also delve into a fur
ther use case on queer fairness. Along the way\, I'll try to connect to th
e work of other organistions\, like the AI and Shared Prosperity initiativ
e by the Partnership on AI\, and Royal Society programme on digital techno
logies for the planet.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/13/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Chris Maddison (University of Toronto and Vector Institute)
DTSTART:20210111T203000Z
DTEND:20210111T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/14
DESCRIPTION:Title: Gradient Estimation with Stochastic Softmax Tricks\nb
y Chris Maddison (University of Toronto and Vector Institute) as part of M
achine Learning Advances and Applications Seminar\n\n\nAbstract\nGradient
estimation is an important problem in modern machine learning frameworks t
hat rely heavily on gradient-based optimization. For gradient estimation i
n the presence of discrete random variables\, the Gumbel-based relaxed gra
dient estimators are easy to implement and low variance\, but the goal of
scaling them comprehensively to large combinatorial distributions is still
outstanding. Working within the perturbation model framework\, we introdu
ce stochastic softmax tricks\, which generalize the Gumbel-Softmax trick t
o combinatorial spaces. Our framework is a unified perspective on existing
relaxed estimators for perturbation models\, and it contains many novel r
elaxations. We design structured relaxations for subset selection\, spanni
ng trees\, arborescences\, and others. We consider an application to helpi
ng make machine learning models more explainable.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/14/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Courtney Paquette (McGill University)
DTSTART:20211108T200000Z
DTEND:20211108T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/15
DESCRIPTION:Title: SGD in the Large: Average-case Analysis\, Asymptotics\, a
nd Stepsize Criticality\nby Courtney Paquette (McGill University) as p
art of Machine Learning Advances and Applications Seminar\n\n\nAbstract\nI
n this talk\, I will present a framework\, inspired by random matrix theor
y\, for analyzing the dynamics of stochastic gradient descent (SGD) when b
oth the number of samples and dimensions are large. Using this new framewo
rk\, we show that the dynamics of SGD on a least squares problem with rand
om data become deterministic in the large sample and dimensional limit. Fu
rthermore\, the limiting dynamics are governed by a Volterra integral equa
tion. This model predicts that SGD undergoes a phase transition at an expl
icitly given critical stepsize that ultimately affects its convergence rat
e\, which we also verify experimentally. Finally\, when input data is isot
ropic\, we provide explicit expressions for the dynamics and average-case
convergence rates. These rates show significant improvement over the worst
-case complexities.\n\nBio: Courtney Paquette is an assistant professor at
McGill University and a CIFAR Canada AI chair\, MILA. Paquette's research
broadly focuses on designing and analyzing algorithms for large-scale opt
imization problems\, motivated by applications in data science. She receiv
ed her PhD from the mathematics department at the University of Washington
(2017)\, held postdoctoral positions at Lehigh University (2017-2018) and
University of Waterloo (NSF postdoctoral fellowship\, 2018-2019)\, and wa
s a research scientist at Google Research\, Brain Montreal (2019-2020).\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/15/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Lester Mackey (Microsoft/Stanford University)
DTSTART:20211122T200000Z
DTEND:20211122T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/16
DESCRIPTION:Title: Kernel Thinning and Stein Thinning\nby Lester Mackey
(Microsoft/Stanford University) as part of Machine Learning Advances and A
pplications Seminar\n\n\nAbstract\nThis talk will introduce two new tools
for summarizing a probability distribution more effectively than independe
nt sampling or standard Markov chain Monte Carlo thinning:\n\n1. Given an
initial n point summary (for example\, from independent sampling or a Mark
ov chain)\, kernel thinning finds a subset of only square-root n points wi
th comparable worst-case integration error across a reproducing kernel Hil
bert space.\n\n2. If the initial summary suffers from biases due to off-ta
rget sampling\, tempering\, or burn-in\, Stein thinning simultaneously com
presses the summary and improves the accuracy by correcting for these bias
es.\n\nThese tools are especially well-suited for tasks that incur substan
tial downstream computation costs per summary point like organ and tissue
modeling in which each simulation consumes 1000s of CPU hours.\n\nBio: Les
ter Mackey is a Principal Researcher at Microsoft Research\, where he deve
lops machine learning methods\, models\, and theory for large-scale learni
ng tasks driven by applications from climate forecasting\, healthcare\, an
d the social good. Lester moved to Microsoft from Stanford University\, wh
ere he was an assistant professor of Statistics and (by courtesy) of Compu
ter Science. He earned his PhD in Computer Science and MA in Statistics fr
om UC Berkeley and his BSE in Computer Science from Princeton University.
He co-organized the second place team in the Netflix Prize competition for
collaborative filtering\, won the Prize4Life ALS disease progression pred
iction challenge\, won prizes for temperature and precipitation forecastin
g in the yearlong real-time Subseasonal Climate Forecast Rodeo\, and recei
ved best paper and best student paper awards from the ACM Conference on Pr
ogramming Language Design and Implementation and the International Confere
nce on Machine Learning.\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/16/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Andreas Krause (ETH Zürich)
DTSTART:20220124T200000Z
DTEND:20220124T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/17
DESCRIPTION:by Andreas Krause (ETH Zürich) as part of Machine Learning Ad
vances and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/17/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Joan Bruna Estrach (New York University)
DTSTART:20220207T200000Z
DTEND:20220207T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/18
DESCRIPTION:by Joan Bruna Estrach (New York University) as part of Machine
Learning Advances and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/18/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Matthew Johnson (Google)
DTSTART:20220214T200000Z
DTEND:20220214T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/19
DESCRIPTION:by Matthew Johnson (Google) as part of Machine Learning Advanc
es and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/19/
END:VEVENT
BEGIN:VEVENT
SUMMARY:David Pfau (DeepMind)
DTSTART:20220307T200000Z
DTEND:20220307T210000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/20
DESCRIPTION:by David Pfau (DeepMind) as part of Machine Learning Advances
and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/20/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Jan Peters (Technische Universitaet Darmstadt)
DTSTART:20220321T190000Z
DTEND:20220321T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/21
DESCRIPTION:by Jan Peters (Technische Universitaet Darmstadt) as part of M
achine Learning Advances and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/21/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Philipp Hennig (Max Planck Institute for Intelligent Systems)
DTSTART:20220404T190000Z
DTEND:20220404T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/22
DESCRIPTION:by Philipp Hennig (Max Planck Institute for Intelligent System
s) as part of Machine Learning Advances and Applications Seminar\n\nAbstra
ct: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/22/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Cynthia Rudin (Duke University)
DTSTART:20220425T190000Z
DTEND:20220425T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/23
DESCRIPTION:by Cynthia Rudin (Duke University) as part of Machine Learning
Advances and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/23/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Anima Anandkumar (California Institute of Technology)
DTSTART:20220502T190000Z
DTEND:20220502T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/24
DESCRIPTION:by Anima Anandkumar (California Institute of Technology) as pa
rt of Machine Learning Advances and Applications Seminar\n\nAbstract: TBA\
n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/24/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Radford Neal (University of Toronto)
DTSTART:20220516T190000Z
DTEND:20220516T200000Z
DTSTAMP:20260404T094205Z
UID:Machine_Learning/25
DESCRIPTION:by Radford Neal (University of Toronto) as part of Machine Lea
rning Advances and Applications Seminar\n\nAbstract: TBA\n
LOCATION:https://stable.researchseminars.org/talk/Machine_Learning/25/
END:VEVENT
END:VCALENDAR