Skip to main content

Publications

Learning grammar with a divide-and-concur neural network

Cornell Affiliated Author(s)
Author
S. Deyo
V. Elser
Abstract

We implement a divide-and-concur iterative projection approach to context-free grammar inference. Unlike most state-of-the-art models of natural language processing, our method requires a relatively small number of discrete parameters, making the inferred grammar directly interpretable - one can read off from a solution how to construct grammatically valid sentences. Another advantage of our approach is the ability to infer meaningful grammatical rules from just a few sentences, compared to the hundreds of gigabytes of training data many other models employ.

Journal
Physical Review E
Date Published
Group (Lab)
Veit Elser Group

AVOIDING TRAPS IN NONCONVEX PROBLEMS

Cornell Affiliated Author(s)
Author
S. Deyo
V. Elser
Abstract

Iterative projection methods may become trapped at non-solutions when the constraint sets are nonconvex. Two kinds of parameters are available to help avoid this behavior and this study gives examples of both. The first kind of parameter, called a hyperparameter, includes any kind of parameter that appears in the definition of the iteration rule itself. The second kind comprises metric parameters in the definition of the constraint sets, a feature that arises when the problem to be solved has two or more kinds of variables.

Journal
Journal of Applied and Numerical Optimization
Date Published
Group (Lab)
Veit Elser Group

Learning without loss

Cornell Affiliated Author(s)
Author
V. Elser
Abstract

We explore a new approach for training neural networks where all loss functions are replaced by hard constraints. The same approach is very successful in phase retrieval, where signals are reconstructed from magnitude constraints and general characteristics (sparsity, support, etc.). Instead of taking gradient steps, the optimizer in the constraint based approach, called relaxed–reflect–reflect (RRR), derives its steps from projections to local constraints.

Journal
Fixed Point Theory and Algorithms for Sciences and Engineering
Date Published
Group (Lab)
Veit Elser Group

Charge-order-enhanced capacitance in semiconductor moiré superlattices

Cornell Affiliated Author(s)
Author
T. Li
J. Zhu
Y. Tang
K. Watanabe
T. Taniguchi
V. Elser
J. Shan
K.F. Mak
Abstract

Van der Waals moiré materials have emerged as a highly controllable platform to study electronic correlation phenomena1–17. Robust correlated insulating states have recently been discovered at both integer and fractional filling factors of semiconductor moiré systems10–17. In this study we explored the thermodynamic properties of these states by measuring the gate capacitance of MoSe2/WS2 moiré superlattices. We observed a series of incompressible states for filling factors 0–8 and anomalously large capacitance in the intervening compressible regions.

Journal
Nature Nanotechnology
Date Published
Funding Source
FA9550-18-1-0480
DMR-1539918
NNCI-1542081
DE-SC0019481
JPMJCR15F3
Group (Lab)
Jie Shan Group
Kin Fai Mak Group
Veit Elser Group

Glass phenomenology in the hard matrix model

Cornell Affiliated Author(s)
Author
J. Dong
V. Elser
G. Gyawali
K.Y. Jee
J. Kent-Dobias
A. Mandaiya
M. Renz
Y. Su
Abstract

We introduce a new toy model for the study of glasses: the hard-matrix model. This may be viewed as a single particle moving on SO(N), where there is a potential proportional to the one-norm of the matrix. The ground states of the model are 'crystals' where all matrix elements have the same magnitude. These are the Hadamard matrices when N is divisible by four. Just as finding the latter has challenged mathematicians, our model fails to find them upon cooling and instead shows all the behaviors that characterize physical glasses.

Journal
Journal of Statistical Mechanics: Theory and Experiment
Date Published
Group (Lab)
Veit Elser Group

Reconstructing cellular automata rules from observations at nonconsecutive times

Cornell Affiliated Author(s)
Author
V. Elser
Abstract

Recent experiments have shown that a deep neural network can be trained to predict the action of t steps of Conway's Game of Life automaton given millions of examples of this action on random initial states. However, training was never completely successful for t>1, and even when successful, a reconstruction of the elementary rule (t=1) from t>1 data is not within the scope of what the neural network can deliver. We describe an alternative network-like method, based on constraint projections, where this is possible.

Journal
Physical Review E
Date Published
Group (Lab)
Veit Elser Group

Correlated insulating states at fractional fillings of moiré superlattices

Cornell Affiliated Author(s)
Author
Y. Xu
S. Liu
D.A. Rhodes
K. Watanabe
T. Taniguchi
J. Hone
V. Elser
K.F. Mak
J. Shan
Abstract

Quantum particles on a lattice with competing long-range interactions are ubiquitous in physics; transition metal oxides1,2, layered molecular crystals3 and trapped-ion arrays4 are a few examples. In the strongly interacting regime, these systems often show a rich variety of quantum many-body ground states that challenge theory2. The emergence of transition metal dichalcogenide moiré superlattices provides a highly controllable platform in which to study long-range electronic correlations5–12.

Journal
Nature
Date Published
Funding Source
N00014-18-1-2368
DE-SC0019481
JPMJCR15F3
Group (Lab)
Jie Shan Group
Kin Fai Mak Group
Veit Elser Group

Chemistry of the spin- 12 kagome Heisenberg antiferromagnet

Cornell Affiliated Author(s)
Author
Y. Yao
C.J. Umrigar
V. Elser
Abstract

We believe that a necessary first step in understanding the ground-state properties of the spin-12 kagome Heisenberg antiferromagnet is a better understanding of this model's very large number of low-energy singlet states. A description of the low-energy states that is both accurate and amenable for numerical work may ultimately prove to have greater value than knowing only what these properties are, in particular, when they turn on the delicate balance of many small energies.

Journal
Physical Review B
Date Published
Funding Source
ACI-1445606
ACI-1547580
1445606
FA9550-18-1-0095
Group (Lab)
Cyrus Umrigar Group
Veit Elser Group

An enhanced formulation for solving graph coloring problems with the Douglas–Rachford algorithm

Cornell Affiliated Author(s)
Author
F.J. Artacho
R. Campoy
V. Elser
Abstract

We study the behavior of the Douglas–Rachford algorithm on the graph vertex-coloring problem. Given a graph and a number of colors, the goal is to find a coloring of the vertices so that all adjacent vertex pairs have different colors. In spite of the combinatorial nature of this problem, the Douglas–Rachford algorithm was recently shown to be a successful heuristic for solving a wide variety of graph coloring instances, when the problem was cast as a feasibility problem on binary indicator variables. In this work we consider a different formulation, based on semidefinite programming.

Journal
Journal of Global Optimization
Date Published
Group (Lab)
Veit Elser Group