yezhengli-Mr9
commented
3 years ago One of the concerns is that [2] typically has
for fixed subscript i,
while Eq. (1) in [1] shows
for fixed subscript i, and results in
Open yezhengli-Mr9 opened 3 years ago
yezhengli-Mr9
commented
3 years ago One of the concerns is that [2] typically has
for fixed subscript i,
while Eq. (1) in [1] shows
for fixed subscript i, and results in
Hi Colin Graber, As for Eq. (1) (which is also referred later on in Fig. 1, etc) I checked [2,3] for an origin (it does not matter to me when [2,3] majorly discusses the case with parameter w linearly multiplied). Although [2] mentioned various formulations (e.g. SVM_0, SVM_1, SVM_2, etc.) as well as some dual forms, I cannot find the exact counterpart/ origin to Eq. (1) in [1]. I also noticed Eq. (8) in [1] might be heuristically relevant: \rangle \ge \Delta(y_i,y) -\xi_i)
for fixed i.
Can you help me identify the origin of Eq. (1) from [2,3]? Best and thanks, Yezheng
[1] Graber, Colin, and Alexander Schwing. "Graph structured prediction energy networks." Advances in Neural Information Processing Systems. 2019. [2] Tsochantaridis, Ioannis, et al. "Large margin methods for structured and interdependent output variables." Journal of machine learning research 6.Sep (2005): 1453-1484. [3] Taskar, Ben, Carlos Guestrin, and Daphne Koller. "Max-margin Markov networks." Advances in neural information processing systems. 2004.