Differentiation of Blackbox Combinatorial Solvers

D 2020

Differentiation of Blackbox Combinatorial Solvers

VLASTELICA, Marin; Anselm PAULUS; Vít MUSIL; Georg MARTIUS; Michal ROLÍNEK et al.

Základní údaje

Originální název

Differentiation of Blackbox Combinatorial Solvers

Autoři

VLASTELICA, Marin; Anselm PAULUS; Vít MUSIL; Georg MARTIUS a Michal ROLÍNEK

Vydání

8th International Conference on Learning Representations, ICLR 2020, od s. 0-19, 19 s. 2020

Nakladatel

International Conference on Learning Representations, ICLR

Další údaje

Jazyk

angličtina

Typ výsledku

Stať ve sborníku

Obor

10101 Pure mathematics

Utajení

není předmětem státního či obchodního tajemství

Forma vydání

elektronická verze "online"

Odkazy

ICLR, Openreview

Označené pro přenos do RIV

Organizační jednotka

Fakulta informatiky

EID Scopus

2-s2.0-85117864239

Klíčová slova anglicky

Deep learning; Graph theory; Network architecture; Robot programming; Black boxes; Building blockes; Combinatorial algorithm; Combinatorial problem; End to end; Global consistency; Input datas; Multi-object tracking; Neural-networks; Route planning; Traveling salesman problem

Příznaky

Mezinárodní význam, Recenzováno

Změněno: 2. 4. 2025 10:41, RNDr. Pavel Šmerk, Ph.D.

Anotace

V originále

Achieving fusion of deep learning with combinatorial algorithms promises transformative changes to artificial intelligence. One possible approach is to introduce combinatorial building blocks into neural networks. Such end-to-end architectures have the potential to tackle combinatorial problems on raw input data such as ensuring global consistency in multi-object tracking or route planning on maps in robotics. In this work, we present a method that implements an efficient backward pass through blackbox implementations of combinatorial solvers with linear objective functions. We provide both theoretical and experimental backing. In particular, we incorporate the Gurobi MIP solver, Blossom V algorithm, and Dijkstra's algorithm into architectures that extract suitable features from raw inputs for the traveling salesman problem, the min-cost perfect matching problem and the shortest path problem.

Citovat

VLASTELICA, Marin; Anselm PAULUS; Vít MUSIL; Georg MARTIUS a Michal ROLÍNEK. Differentiation of Blackbox Combinatorial Solvers. Online. In 8th International Conference on Learning Representations, ICLR 2020. International Conference on Learning Representations, ICLR, 2020, s. 0-19, 19 s.

@inproceedings{2449803,
   author = {Vlastelica, Marin and Paulus, Anselm and Musil, Vít and Martius, Georg and Rolínek, Michal},
   booktitle = {8th International Conference on Learning Representations, ICLR 2020},
   keywords = {Deep learning; Graph theory; Network architecture; Robot programming; Black boxes; Building blockes; Combinatorial algorithm; Combinatorial problem; End to end; Global consistency; Input datas; Multi-object tracking; Neural-networks; Route planning; Traveling salesman problem},
   howpublished = {elektronická verze "online"},
   language = {eng},
   pages = {0-19},
   publisher = {International Conference on Learning Representations, ICLR},
   title = {Differentiation of Blackbox Combinatorial Solvers},
   url = {https://openreview.net/forum?id=BkevoJSYPB},
   year = {2020}
}

TY  - CONF
ID  - 2449803
AU  - Vlastelica, Marin - Paulus, Anselm - Musil, Vít - Martius, Georg - Rolínek, Michal
PY  - 2020
TI  - Differentiation of Blackbox Combinatorial Solvers
PB  - International Conference on Learning Representations, ICLR
KW  - Deep learning
KW  - Graph theory
KW  - Network architecture
KW  - Robot programming
KW  - Black boxes
KW  - Building blockes
KW  - Combinatorial algorithm
KW  - Combinatorial problem
KW  - End to end
KW  - Global consistency
KW  - Input datas
KW  - Multi-object tracking
KW  - Neural-networks
KW  - Route planning
KW  - Traveling salesman problem
UR  - https://openreview.net/forum?id=BkevoJSYPB
N2  - Achieving fusion of deep learning with combinatorial algorithms promises transformative changes to artificial intelligence. One possible approach is to introduce combinatorial building blocks into neural networks. Such end-to-end architectures have the potential to tackle combinatorial problems on raw input data such as ensuring global consistency in multi-object tracking or route planning on maps in robotics. In this work, we present a method that implements an efficient backward pass through blackbox implementations of combinatorial solvers with linear objective functions. We provide both theoretical and experimental backing. In particular, we incorporate the Gurobi MIP solver, Blossom V algorithm, and Dijkstra's algorithm into architectures that extract suitable features from raw inputs for the traveling salesman problem, the min-cost perfect matching problem and the shortest path problem.
ER  -

VLASTELICA, Marin; Anselm PAULUS; Vít MUSIL; Georg MARTIUS a Michal ROLÍNEK. Differentiation of Blackbox Combinatorial Solvers. Online. In \textit{8th International Conference on Learning Representations, ICLR 2020}. International Conference on Learning Representations, ICLR, 2020, s.~0-19, 19 s.

Přehled o publikaci