Structured Minimally Supervised Learning for Neural Relation Extraction

(1)

Proceedings of NAACL-HLT 2019, pages 3057–3069 3057

Structured Minimally Supervised Learning for Neural

Relation Extraction

Fan Bai and Alan Ritter

Department of Computer Science and Engineering The Ohio State University

Columbus, OH

{bai.313, ritter.1492}@osu.edu

Abstract

We present an approach to minimally su-pervised relation extraction that combines the benefits of learned representations and structured learning, and accurately predicts sentence-level relation mentions given only proposition-level supervision from a KB. By explicitly reasoning about missing data dur-ing learndur-ing, our approach enables large-scale training of 1D convolutional neural networks while mitigating the issue of label noise in-herent in distant supervision. Our approach achieves state-of-the-art results on minimally supervised sentential relation extraction, out-performing a number of baselines, including a competitive approach that uses the attention

layer of a purely neural model.1

1 Introduction

Recent years have seen significant progress on tasks such as object detection, automatic speech recognition and machine translation. These per-formance advances are largely driven by the appli-cation of neural network methods on large, high-quality datasets. In contrast, traditional datasets for relation extraction are based on expensive and time-consuming human annotation (Doddington

et al., 2004) and are therefore relatively small.

Distant supervision (Mintz et al., 2009), a

tech-nique which uses existing knowledge bases such as Freebase or Wikipedia as a source of weak su-pervision, enables learning from large quantities of unlabeled text and is a promising approach for scaling up. Recent work has shown promising re-sults from large-scale training of neural networks

for relation extraction (Toutanova et al., 2015;

Zeng et al.,2015).

There are, however, significant challenges due to the inherent noise in distant supervision. For

1 _{Our code and data are publicly available on Github:}

https://github.com/bflashcp3f/PCNN-NMAR

example, Riedel et al. (2010) showed that, when learning using distant supervision from a knowl-edge base, the portion of mis-labeled examples can vary from 13% to 31%. To address this issue,

another line of work has exploredstructured

learn-ing methods that introduce latent variables. An

ex-ample is MultiR (Hoffmann et al.,2011), which is

based on a joint model of relations between enti-ties in a knowledge base and those mentioned in text. This structured learning approach has a num-ber of advantages; for example, by integrating in-ference into the learning procedure it has the po-tential to overcome the challenge of missing facts by ignoring the knowledge base when mention-level classifiers have high confidence (Ritter et al.,

2013;Xu et al., 2013). Prior work on structured

learning from minimal supervision has leveraged sparse feature representations, however, and has therefore not benefited from learned representa-tions, which have recently achieved state-of-the-art results on a broad range of NLP tasks.

In this paper, we present an approach that com-bines the benefits of structured and neural methods for minimally supervised relation extraction. Our proposed model learns sentence representations that are computed by a 1D convolutional neural

network (Collobert et al., 2011) and are used to

define potentials over latent relation mention vari-ables. These mention-level variables are related to observed facts in a KB using a set of determin-istic factors, followed by pairwise potentials that encourage agreement between extracted proposi-tions and observed facts, but also enable inference to override these soft constraints during learning, allowing for the possibility of missing informa-tion. Because marginal inference is intractable in this model, a MAP-based approach to learning is

applied (Taskar et al.,2004).

(2)

representa-tions, including Structured Prediction Energy

Net-works (SPENs) (Belanger and McCallum,2016);

the key differences are the application to mini-mally supervised relation extraction and the inclu-sion of latent variables with deterministic factors, which we demonstrate enables effective learning in the presence of missing data in distant supervi-sion. Our proposed method achieves state-of-the-art results on minimally supervised sentential rela-tion extracrela-tion, outperforming a number of base-lines including one that leverages the attention

layer of a purely neural model (Lin et al.,2016).

2 A Latent Variable Model for Neural

Relation Extraction

In this section we present our model, which com-bines continuous representations with structured

learning. We first review the problem setting

and introduce notation, next we present our ap-proach to extracting feature representations which is based on the piecewise convolutional neural net-work (PCNN) model of Zeng et. al. (2015) and includes positional embeddings (Collobert et al., 2011). Finally we describe how this can be com-bined with structured latent variable models that reason about overlapping relations and missing data during learning.

2.1 Assumptions and Problem Formulation

Given a set of sentences, s = s1, s2. . . , sn that

mention a pair of knowledge base entitiese1 and

e2(dyad), our goal is to predict which relation,r,

is mentioned betweene1 ande2 in the context of

each sentence, represented by a set of hidden

vari-ables, z = z1, z2, . . . zn. Relations are selected

from a fixed set drawn from a knowledge base,

in addition toNA(no relation). Minimally

super-vised learning is more difficult than supersuper-vised re-lation extraction, because we do not have direct access to relation labels on the training sentences. Instead, during learning, we are only provided with information about what relations hold

be-tweene1ande2according to the KB. The problem

is further complicated by the fact that most KBs are highly incomplete (this is the reason we want to extend them by extracting information from text in the first place), which effectively leads to false-negatives during learning. Furthermore, there are many overlapping relations between dyads, so it is easy for a model trained using minimal supervi-sion from a KB to confuse these relationships. All

of these issues are addressed to some degree by the structured learning approach that we present

in Section2.3. First, however we present our

ap-proach to feature representation based on convolu-tional neural networks.

t

j

<latexit sha1_base64="rj6OOVsYR0YK/1stMvAydejdR1U=">AAACFHicbVBNS8NAEN34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdiVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fiq4Bsf5tlZW19Y3Nitb9vbO7t5+9eDwSSeZosyliUhUxyeaCS6ZCxwE66SKkdgXrO0Pbwu9/cKU5ol8hFHKvJhEkoecEjDUA/Sf+9WaU3emhZdBowQ1VFarX/3pBQnNYiaBCqJ1t+Gk4OVEAaeCje1epllK6JBErGugJDHTXj61OsanhglwmCjTEvCU/buRk1jrUeybyZjAQC9qBfmf1s0gvPJyLtMMmKSzQ2EmMCS4+BsHXDEKYmQAoYobr5gOiCIUTDr23JlAF97mHsnTKDSmx7Ztm7wai+ksA/e8fl137i9qzZsyuAo6RifoDDXQJWqiO9RCLqIoQq/oDU2sifVufVifs9EVq9w5QnNlff0CZ2qeyA==</latexit><latexit sha1_base64="rj6OOVsYR0YK/1stMvAydejdR1U=">AAACFHicbVBNS8NAEN34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdiVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fiq4Bsf5tlZW19Y3Nitb9vbO7t5+9eDwSSeZosyliUhUxyeaCS6ZCxwE66SKkdgXrO0Pbwu9/cKU5ol8hFHKvJhEkoecEjDUA/Sf+9WaU3emhZdBowQ1VFarX/3pBQnNYiaBCqJ1t+Gk4OVEAaeCje1epllK6JBErGugJDHTXj61OsanhglwmCjTEvCU/buRk1jrUeybyZjAQC9qBfmf1s0gvPJyLtMMmKSzQ2EmMCS4+BsHXDEKYmQAoYobr5gOiCIUTDr23JlAF97mHsnTKDSmx7Ztm7wai+ksA/e8fl137i9qzZsyuAo6RifoDDXQJWqiO9RCLqIoQq/oDU2sifVufVifs9EVq9w5QnNlff0CZ2qeyA==</latexit><latexit sha1_base64="rj6OOVsYR0YK/1stMvAydejdR1U=">AAACFHicbVBNS8NAEN34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdiVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fiq4Bsf5tlZW19Y3Nitb9vbO7t5+9eDwSSeZosyliUhUxyeaCS6ZCxwE66SKkdgXrO0Pbwu9/cKU5ol8hFHKvJhEkoecEjDUA/Sf+9WaU3emhZdBowQ1VFarX/3pBQnNYiaBCqJ1t+Gk4OVEAaeCje1epllK6JBErGugJDHTXj61OsanhglwmCjTEvCU/buRk1jrUeybyZjAQC9qBfmf1s0gvPJyLtMMmKSzQ2EmMCS4+BsHXDEKYmQAoYobr5gOiCIUTDr23JlAF97mHsnTKDSmx7Ztm7wai+ksA/e8fl137i9qzZsyuAo6RifoDDXQJWqiO9RCLqIoQq/oDU2sifVufVifs9EVq9w5QnNlff0CZ2qeyA==</latexit>

zi

<latexit sha1_base64="/wZpS6uWRmtWnxFSSYnO5XPc+Wo=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl262YTdjVBDf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7WcqFUvEox6l1ItxJFjICNaGenjps3615tSdaaFl0ChBDcpq9as/vSAhWUyFJhwr1W04qfZyLDUjnI7tXqZoiskQR7RroMAxVV4+tTpGp4YJUJhI00KjKft3I8exUqPYN5Mx1gO1qBXkf1o30+GVlzORZpoKMjsUZhzpBBV/o4BJSjQfGYCJZMYrIgMsMdEmHXvuTKAKb3OP5GkUGtNj27ZNXo3FdJaBe16/rjv3F7XmTRlcBY7hBM6gAZfQhDtogQsEIniFN5hYE+vd+rA+Z6MrVrlzBHNlff0Cb9yezQ==</latexit><latexit sha1_base64="/wZpS6uWRmtWnxFSSYnO5XPc+Wo=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl262YTdjVBDf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7WcqFUvEox6l1ItxJFjICNaGenjps3615tSdaaFl0ChBDcpq9as/vSAhWUyFJhwr1W04qfZyLDUjnI7tXqZoiskQR7RroMAxVV4+tTpGp4YJUJhI00KjKft3I8exUqPYN5Mx1gO1qBXkf1o30+GVlzORZpoKMjsUZhzpBBV/o4BJSjQfGYCJZMYrIgMsMdEmHXvuTKAKb3OP5GkUGtNj27ZNXo3FdJaBe16/rjv3F7XmTRlcBY7hBM6gAZfQhDtogQsEIniFN5hYE+vd+rA+Z6MrVrlzBHNlff0Cb9yezQ==</latexit><latexit sha1_base64="/wZpS6uWRmtWnxFSSYnO5XPc+Wo=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl262YTdjVBDf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7WcqFUvEox6l1ItxJFjICNaGenjps3615tSdaaFl0ChBDcpq9as/vSAhWUyFJhwr1W04qfZyLDUjnI7tXqZoiskQR7RroMAxVV4+tTpGp4YJUJhI00KjKft3I8exUqPYN5Mx1gO1qBXkf1o30+GVlzORZpoKMjsUZhzpBBV/o4BJSjQfGYCJZMYrIgMsMdEmHXvuTKAKb3OP5GkUGtNj27ZNXo3FdJaBe16/rjv3F7XmTRlcBY7hBM6gAZfQhDtogQsEIniFN5hYE+vd+rA+Z6MrVrlzBHNlff0Cb9yezQ==</latexit>

…

dj

<latexit sha1_base64="l8LqiwKV7DcUrWNQPgYWQFtTCwk=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdjVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7RcqFUvEox6l1ItxJFjICNaGegj6z/1qzak700LLoFGCGpTV6ld/ekFCspgKTThWqttwUu3lWGpGOB3bvUzRFJMhjmjXQIFjqrx8anWMTg0ToDCRpoVGU/bvRo5jpUaxbyZjrAdqUSvI/7RupsMrL2cizTQVZHYozDjSCSr+RgGTlGg+MgATyYxXRAZYYqJNOvbcmUAV3uYeydMoNKbHtm2bvBqL6SwD97x+XXfuL2rNmzK4ChzDCZxBAy6hCXfQAhcIRPAKbzCxJta79WF9zkZXrHLnCObK+voFTGqeuA==</latexit><latexit sha1_base64="l8LqiwKV7DcUrWNQPgYWQFtTCwk=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdjVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7RcqFUvEox6l1ItxJFjICNaGegj6z/1qzak700LLoFGCGpTV6ld/ekFCspgKTThWqttwUu3lWGpGOB3bvUzRFJMhjmjXQIFjqrx8anWMTg0ToDCRpoVGU/bvRo5jpUaxbyZjrAdqUSvI/7RupsMrL2cizTQVZHYozDjSCSr+RgGTlGg+MgATyYxXRAZYYqJNOvbcmUAV3uYeydMoNKbHtm2bvBqL6SwD97x+XXfuL2rNmzK4ChzDCZxBAy6hCXfQAhcIRPAKbzCxJta79WF9zkZXrHLnCObK+voFTGqeuA==</latexit><latexit sha1_base64="l8LqiwKV7DcUrWNQPgYWQFtTCwk=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KqkI6q3oxWNFYwttKJvNJl272YTdjVBCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5fsqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahLEp7Ijo8V5UxQVzPNaSeVFMc+p21/eFvo7RcqFUvEox6l1ItxJFjICNaGegj6z/1qzak700LLoFGCGpTV6ld/ekFCspgKTThWqttwUu3lWGpGOB3bvUzRFJMhjmjXQIFjqrx8anWMTg0ToDCRpoVGU/bvRo5jpUaxbyZjrAdqUSvI/7RupsMrL2cizTQVZHYozDjSCSr+RgGTlGg+MgATyYxXRAZYYqJNOvbcmUAV3uYeydMoNKbHtm2bvBqL6SwD97x+XXfuL2rNmzK4ChzDCZxBAy6hCXfQAhcIRPAKbzCxJta79WF9zkZXrHLnCObK+voFTGqeuA==</latexit>

… … … …

Sentence Representation:

Input Sentence:

Relation Observed in Database: (Observed during learning)

Relation Mentioned in Corpus: (Latent)

Embedding Layer: PCNN

Relation Mentioned in S: (Latent)

E

_⇥

E

<latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="FViP8FQW0xxf/P1rHASRXeO3OD4=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM1m0y7dbMLuS6GE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EuRQGPe/bqe3tHxwe1Y/dk4Z7enbebLyZrNCMByyTme5F1HApFA9QoOS9XHOaRpJ3o8lT5XenXBuRqVec5TxM6UiJRDCKVuoMmy2v7S1Bdom/Ji1YY9j8GcQZK1KukElqTN/3cgxLqlEwyefuoDA8p2xCR7xvqaIpN2G5jDkn11aJSZJpOwrJUv37o6SpMbM0si9TimOz7VXif16/wOQ+LIXKC+SKrRYlhSSYkepmEgvNGcqZJZRpYbMSNqaaMrTNuBtrYlNl2zikzEeJDT13XdfW5W+Xs0uC2/ZD23vxoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AihkmwE=</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="GbxTenL2pzppd9iPWzumUf/ZTNg=">AAACG3icbVBNS8NAEJ34WeNX1aOXxSJ4KokX9VaUgscKxhbbUDabTbt0swm7G6GE/guv9td4Eq8e/DGCmzYH2zow8Hgzw3vzgpQzpR3n21pb39jc2q7s2Lt7+weH1aPjJ5VkklCPJDyRnQArypmgnmaa004qKY4DTtvB6K6Yt1+oVCwRj3qcUj/GA8EiRrA21HMT9TSLqULNfrXm1J1ZoVXglqAGZbX61Z9emJAspkITjpXquk6q/RxLzQinE7uXKZpiMsID2jVQYCPj5zPHE3RumBBFiTQtNJqxfy9yHCs1jgOzGWM9VMuzgvxv1s10dO3nTKSZpoLMhaKMI52g4n0UMkmJ5mMDMJHMeEVkiCUm2oRkL8iEqvC28EieDiJjemLbtsnLXU5nFXiX9Zu68+DUGrdlcBU4hTO4ABeuoAH30AIPCAh4hTeYWlPr3fqwPuera1Z5cwILZX39AuXgoRU=</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="FViP8FQW0xxf/P1rHASRXeO3OD4=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM1m0y7dbMLuS6GE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EuRQGPe/bqe3tHxwe1Y/dk4Z7enbebLyZrNCMByyTme5F1HApFA9QoOS9XHOaRpJ3o8lT5XenXBuRqVec5TxM6UiJRDCKVuoMmy2v7S1Bdom/Ji1YY9j8GcQZK1KukElqTN/3cgxLqlEwyefuoDA8p2xCR7xvqaIpN2G5jDkn11aJSZJpOwrJUv37o6SpMbM0si9TimOz7VXif16/wOQ+LIXKC+SKrRYlhSSYkepmEgvNGcqZJZRpYbMSNqaaMrTNuBtrYlNl2zikzEeJDT13XdfW5W+Xs0uC2/ZD23vxoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AihkmwE=</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="GbxTenL2pzppd9iPWzumUf/ZTNg=">AAACG3icbVBNS8NAEJ34WeNX1aOXxSJ4KokX9VaUgscKxhbbUDabTbt0swm7G6GE/guv9td4Eq8e/DGCmzYH2zow8Hgzw3vzgpQzpR3n21pb39jc2q7s2Lt7+weH1aPjJ5VkklCPJDyRnQArypmgnmaa004qKY4DTtvB6K6Yt1+oVCwRj3qcUj/GA8EiRrA21HMT9TSLqULNfrXm1J1ZoVXglqAGZbX61Z9emJAspkITjpXquk6q/RxLzQinE7uXKZpiMsID2jVQYCPj5zPHE3RumBBFiTQtNJqxfy9yHCs1jgOzGWM9VMuzgvxv1s10dO3nTKSZpoLMhaKMI52g4n0UMkmJ5mMDMJHMeEVkiCUm2oRkL8iEqvC28EieDiJjemLbtsnLXU5nFXiX9Zu68+DUGrdlcBU4hTO4ABeuoAH30AIPCAh4hTeYWlPr3fqwPuera1Z5cwILZX39AuXgoRU=</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="FViP8FQW0xxf/P1rHASRXeO3OD4=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM1m0y7dbMLuS6GE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EuRQGPe/bqe3tHxwe1Y/dk4Z7enbebLyZrNCMByyTme5F1HApFA9QoOS9XHOaRpJ3o8lT5XenXBuRqVec5TxM6UiJRDCKVuoMmy2v7S1Bdom/Ji1YY9j8GcQZK1KukElqTN/3cgxLqlEwyefuoDA8p2xCR7xvqaIpN2G5jDkn11aJSZJpOwrJUv37o6SpMbM0si9TimOz7VXif16/wOQ+LIXKC+SKrRYlhSSYkepmEgvNGcqZJZRpYbMSNqaaMrTNuBtrYlNl2zikzEeJDT13XdfW5W+Xs0uC2/ZD23vxoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AihkmwE=</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="+yFGraS7BvRuu1bzJ/aUoWNT4bo=">AAACEHicbVDLSsNAFL2prxqrVrduBovgqiRu1J0gBZcVjC22pUwmk3boZBJmboQS+hdu7de4En/AjxGcPha29cKFwzkznHNPmElh0PO+ndLW9s7uXnnfPagcHh1XTyrPJs014wFLZarbITVcCsUDFCh5O9OcJqHkrXB0P9Nbr1wbkaonHGe8l9CBErFgFC310iBdFAk3pNGv1ry6Nx+yCfwlqMFymv3qTzdKWZ5whUxSYzq+l2GvoBoFk3zidnPDM8pGdMA7FipqbXrFPPGEXFgmInGq7Sokc/bvj4ImxoyT0L5MKA7NujYj/9M6OcY3vUKoLEeu2MIoziXBlMzOJ5HQnKEcW0CZFjYrYUOqKUNbkrtiE5lZtpVDimwQ29AT13VtX/56O5sguKrf1r1HD8pwBudwCT5cwx08QBMCYKDgDd5h6kydD+dzUWzJWTZ8CivjfP0C+ICfkw==</latexit><latexit sha1_base64="GbxTenL2pzppd9iPWzumUf/ZTNg=">AAACG3icbVBNS8NAEJ34WeNX1aOXxSJ4KokX9VaUgscKxhbbUDabTbt0swm7G6GE/guv9td4Eq8e/DGCmzYH2zow8Hgzw3vzgpQzpR3n21pb39jc2q7s2Lt7+weH1aPjJ5VkklCPJDyRnQArypmgnmaa004qKY4DTtvB6K6Yt1+oVCwRj3qcUj/GA8EiRrA21HMT9TSLqULNfrXm1J1ZoVXglqAGZbX61Z9emJAspkITjpXquk6q/RxLzQinE7uXKZpiMsID2jVQYCPj5zPHE3RumBBFiTQtNJqxfy9yHCs1jgOzGWM9VMuzgvxv1s10dO3nTKSZpoLMhaKMI52g4n0UMkmJ5mMDMJHMeEVkiCUm2oRkL8iEqvC28EieDiJjemLbtsnLXU5nFXiX9Zu68+DUGrdlcBU4hTO4ABeuoAH30AIPCAh4hTeYWlPr3fqwPuera1Z5cwILZX39AuXgoRU=</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit><latexit sha1_base64="9n6P/Psi0wW6toZoaWyZIG9f3WQ=">AAACG3icbVBNS8NAEN3Urxq/qh69LBbBU0lEUG9FKXisYG2xDWWz2bRLN5uwOxFK6L/wan+NJ/HqwR8juGlzsK0DA483M7w3z08E1+A431ZpbX1jc6u8be/s7u0fVA6PnnScKspaNBax6vhEM8ElawEHwTqJYiTyBWv7o7t83n5hSvNYPsI4YV5EBpKHnBIw1HMD94BHTONGv1J1as6s8CpwC1BFRTX7lZ9eENM0YhKoIFp3XScBLyMKOBVsYvdSzRJCR2TAugZKYmS8bOZ4gs8ME+AwVqYl4Bn79yIjkdbjyDebEYGhXp7l5H+zbgrhtZdxmaTAJJ0LhanAEOP8fRxwxSiIsQGEKm68YjokilAwIdkLMoHOvS08kiWD0Jie2LZt8nKX01kFrYvaTc15uKzWb4vgyugEnaJz5KIrVEf3qIlaiCKJXtEbmlpT6936sD7nqyWruDlGC2V9/QLnIKEZ</latexit>

R

<latexit sha1_base64="PWpwglb+88EzfT2G1nc76tNdtV0=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWMrxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGRLZ3J</latexit><latexit sha1_base64="PWpwglb+88EzfT2G1nc76tNdtV0=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWMrxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGRLZ3J</latexit><latexit sha1_base64="PWpwglb+88EzfT2G1nc76tNdtV0=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWMrxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGRLZ3J</latexit>

S

<latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="FViP8FQW0xxf/P1rHASRXeO3OD4=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM1m0y7dbMLuS6GE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EuRQGPe/bqe3tHxwe1Y/dk4Z7enbebLyZrNCMByyTme5F1HApFA9QoOS9XHOaRpJ3o8lT5XenXBuRqVec5TxM6UiJRDCKVuoMmy2v7S1Bdom/Ji1YY9j8GcQZK1KukElqTN/3cgxLqlEwyefuoDA8p2xCR7xvqaIpN2G5jDkn11aJSZJpOwrJUv37o6SpMbM0si9TimOz7VXif16/wOQ+LIXKC+SKrRYlhSSYkepmEgvNGcqZJZRpYbMSNqaaMrTNuBtrYlNl2zikzEeJDT13XdfW5W+Xs0uC2/ZD23vxoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AihkmwE=</latexit><latexit sha1_base64="90fHziq4dhyl/jBTfSyrIqgn4G8=">AAACB3icbVBNS8NAFHzxs8aq1auXxSJ4KqkX9SZ48diisYU2lM3mpV262YTdjVBCf4FX+2s8ib/CHyO4aXuwrQMPhpld3rwJM8G18bxvZ2t7Z3dvv3LgHlaPjk9qp9UXneaKoc9SkapuSDUKLtE33AjsZgppEgrshOOH0u+8otI8lc9mkmGQ0KHkMWfUWKn9NKjVvYY3B9kkzSWpwxKtQe2nH6UsT1AaJqjWvaaXmaCgynAmcOr2c40ZZWM6xJ6lkiaog2IedEourRKROFV2pCFz9e+PgiZaT5LQvkyoGel1rxT/83q5iW+DgsssNyjZYlGcC2JSUl5NIq6QGTGxhDLFbVbCRlRRZmw37sqaSJfZVg4psmFsQ09d17V9Ndfb2ST+deOu4bU9qMA5XMAVNOEG7uERWuADA4Q3eIeZM3M+nM9FsVvOsuEzWIHz9QvHApxN</latexit><latexit sha1_base64="90fHziq4dhyl/jBTfSyrIqgn4G8=">AAACB3icbVBNS8NAFHzxs8aq1auXxSJ4KqkX9SZ48diisYU2lM3mpV262YTdjVBCf4FX+2s8ib/CHyO4aXuwrQMPhpld3rwJM8G18bxvZ2t7Z3dvv3LgHlaPjk9qp9UXneaKoc9SkapuSDUKLtE33AjsZgppEgrshOOH0u+8otI8lc9mkmGQ0KHkMWfUWKn9NKjVvYY3B9kkzSWpwxKtQe2nH6UsT1AaJqjWvaaXmaCgynAmcOr2c40ZZWM6xJ6lkiaog2IedEourRKROFV2pCFz9e+PgiZaT5LQvkyoGel1rxT/83q5iW+DgsssNyjZYlGcC2JSUl5NIq6QGTGxhDLFbVbCRlRRZmw37sqaSJfZVg4psmFsQ09d17V9Ndfb2ST+deOu4bU9qMA5XMAVNOEG7uERWuADA4Q3eIeZM3M+nM9FsVvOsuEzWIHz9QvHApxN</latexit><latexit sha1_base64="ka6nYB2DGMai3DDXv6eY12IqDdQ=">AAACEnicbVBNS8NAEN3Urxq/qh69LBbBU0m8qLeiF48tGltoQ9lsJu3SzSbsboQS+gu82l/jSbz6B/wxgps2B9s6MPB4b4Z584KUM6Ud59uqbGxube9Ud+29/YPDo9rxybNKMknBowlPZDcgCjgT4GmmOXRTCSQOOHSC8X2hd15AKpaIJz1JwY/JULCIUaIN1X4c1OpOw5kXXgduCeqorNag9tMPE5rFIDTlRKme66Taz4nUjHKY2v1MQUromAyhZ6AgMSg/nxud4gvDhDhKpGmh8Zz9u5GTWKlJHJjJmOiRWtUK8j+tl+noxs+ZSDMNgi4ORRnHOsHF1zhkEqjmEwMIlcx4xXREJKHaZGMvnQlV4W3pkTwdRsb01LZtk5e7ms468K4atw2n7dSbd2VwVXSGztElctE1aqIH1EIeogjQK3pDM2tmvVsf1uditGKVO6doqayvX5GbncY=</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit><latexit sha1_base64="VxdDw3i95zzktmeiV75BR6La8YE=">AAACEnicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWOLxhbaUDabTbt0swm7G6GE/gKv9td4Eq/+AX+M4KbNwbYODDzem2HevCDlTGnH+bbW1jc2t7YrO/bu3v7BYfXo+FklmSTUIwlPZCfAinImqKeZ5rSTSorjgNN2MLov9PYLlYol4kmPU+rHeCBYxAjWhmo99qs1p+7MCq0CtwQ1KKvZr/70woRkMRWacKxU13VS7edYakY4ndi9TNEUkxEe0K6BAsdU+fnM6ASdGyZEUSJNC41m7N+NHMdKjePATMZYD9WyVpD/ad1MRzd+zkSaaSrI/FCUcaQTVHyNQiYp0XxsACaSGa+IDLHERJts7IUzoSq8LTySp4PImJ7Ytm3ycpfTWQXeZf227rSuao27MrgKnMIZXIAL19CAB2iCBwQovMIbTK2p9W59WJ/z0TWr3DmBhbK+fgGS253K</latexit>

x

i

<latexit sha1_base64="Ljx76w/zKSbzSgwpKCF0+btedDg=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVhCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSY71qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CbM6ezA==</latexit><latexit sha1_base64="Ljx76w/zKSbzSgwpKCF0+btedDg=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVhCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSY71qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CbM6ezA==</latexit><latexit sha1_base64="Ljx76w/zKSbzSgwpKCF0+btedDg=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVhCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSY71qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CbM6ezA==</latexit><latexit sha1_base64="pgt7e/aUqPh6UaONFIne6e+X0Ds=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM3mpV262YTdTaGE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EueDaeN63U9vbPzg8qh+7Jw339Oy82XjTWaEYBiwTmepFVKPgEgPDjcBerpCmkcBuNHmq/O4UleaZfDWzHMOUjiRPOKPGSp1hs+W1vSXILvHXpAVrDJs/gzhjRYrSMEG17vtebsKSKsOZwLk7KDTmlE3oCPuWSpqiDstlzDm5tkpMkkzZkYYs1b8/SppqPUsj+zKlZqy3vUr8z+sXJrkPSy7zwqBkq0VJIYjJSHUziblCZsTMEsoUt1kJG1NFmbHNuBtrYl1l2zikzEeJDT13XdfW5W+Xs0uC2/ZD23/xoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AiipmwI=</latexit><latexit sha1_base64="2MIf+OioKeGR9svToQuAG0gdjSM=">AAACCXicbVDNSsNAGPzib41Vq1cvi0XwVBIv6k3w4rGisYU2lM1mky7d3YTdjVhCH8GrfRpP4kP4MIKbtgfbOvDBMLPLN99EOWfaeN63s7G5tb2zW9tz9+sHh0eN4/qzzgpFaEAynqluhDXlTNLAMMNpN1cUi4jTTjS6q/zOC1WaZfLJjHMaCpxKljCCjZUeXwds0Gh6LW8GtE78BWnCAu1B46cfZ6QQVBrCsdY938tNWGJlGOF04vYLTXNMRjilPUslFlSH5SzqBJ1bJUZJpuxIg2bq3x8lFlqPRWRfCmyGetWrxP+8XmGS67BkMi8MlWS+KCk4Mhmq7kYxU5QYPrYEE8VsVkSGWGFibDvu0ppYV9mWDinzNLGhJ67r2r781XbWSXDZumn5Dx7U4BTO4AJ8uIJbuIc2BEAghTd4h6kzdT6cz3mxG86i4RNYgvP1C5Y7nU0=</latexit>

<latexit sha1_base64="2MIf+OioKeGR9svToQuAG0gdjSM=">AAACCXicbVDNSsNAGPzib41Vq1cvi0XwVBIv6k3w4rGisYU2lM1mky7d3YTdjVhCH8GrfRpP4kP4MIKbtgfbOvDBMLPLN99EOWfaeN63s7G5tb2zW9tz9+sHh0eN4/qzzgpFaEAynqluhDXlTNLAMMNpN1cUi4jTTjS6q/zOC1WaZfLJjHMaCpxKljCCjZUeXwds0Gh6LW8GtE78BWnCAu1B46cfZ6QQVBrCsdY938tNWGJlGOF04vYLTXNMRjilPUslFlSH5SzqBJ1bJUZJpuxIg2bq3x8lFlqPRWRfCmyGetWrxP+8XmGS67BkMi8MlWS+KCk4Mhmq7kYxU5QYPrYEE8VsVkSGWGFibDvu0ppYV9mWDinzNLGhJ67r2r781XbWSXDZumn5Dx7U4BTO4AJ8uIJbuIc2BEAghTd4h6kzdT6cz3mxG86i4RNYgvP1C5Y7nU0=</latexit>

<latexit sha1_base64="7A4bDkgtCs2nSyGVZ28saQ3OwDM=">AAACFHicbVBNS8NAEJ3Urxq/qh69LBbBU0m8qLeiF48VjS20oWw2m3TpZhN2N2IJ/Qle7a/xJF69+2MEt20OtnVg4PHeDPPmBRlnSjvOt1VZW9/Y3Kpu2zu7e/sHtcOjJ5XmklCPpDyVnQArypmgnmaa004mKU4CTtvB8Haqt5+pVCwVj3qUUT/BsWARI1gb6uGlz/q1utNwZoVWgVuCOpTV6td+emFK8oQKTThWqus6mfYLLDUjnI7tXq5ohskQx7RroMAJVX4xszpGZ4YJUZRK00KjGft3o8CJUqMkMJMJ1gO1rE3J/7RurqMrv2AiyzUVZH4oyjnSKZr+jUImKdF8ZAAmkhmviAywxESbdOyFM6Gaelt4pMjiyJge27Zt8nKX01kF3kXjuuHeO/XmTRlcFU7gFM7BhUtowh20wAMCMbzCG0ysifVufVif89GKVe4cw0JZX79rjp7I</latexit>

<latexit sha1_base64="Ljx76w/zKSbzSgwpKCF0+btedDg=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVhCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSY71qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CbM6ezA==</latexit>

si

<latexit sha1_base64="NRg/axUn1QChj5WPHBhm4NsDr3Q=">AAACFHicbVBNS8NAEJ3Urxq/qh69LBbBU0lEUG9FLx4rGltoQ9lsNu3SzSbsboQS+hO82l/jSbx698cIbtocbOvAwOO9GebNC1LOlHacb6uytr6xuVXdtnd29/YPaodHzyrJJKEeSXgiOwFWlDNBPc00p51UUhwHnLaD0V2ht1+oVCwRT3qcUj/GA8EiRrA21KPqs36t7jScWaFV4JagDmW1+rWfXpiQLKZCE46V6rpOqv0cS80IpxO7lymaYjLCA9o1UOCYKj+fWZ2gM8OEKEqkaaHRjP27keNYqXEcmMkY66Fa1gryP62b6ejaz5lIM00FmR+KMo50goq/UcgkJZqPDcBEMuMVkSGWmGiTjr1wJlSFt4VH8nQQGdMT27ZNXu5yOqvAu2jcNNyHy3rztgyuCidwCufgwhU04R5a4AGBAbzCG0ytqfVufVif89GKVe4cw0JZX79kXp7H</latexit>

w

1

<latexit sha1_base64="g1Y4yQGEV9NzThZZTRXwn73mAS4=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSc3vVmlN3poWWgVuCGpTV7FV/umFCspgKTThWquM6qfZzLDUjnI7tbqZoiskQ92nHQIFjqvx8anWMTg0ToiiRpoVGU/bvRo5jpUZxYCZjrAdqUSvI/7ROpqMrP2cizTQVZHYoyjjSCSr+RiGTlGg+MgATyYxXRAZYYqJNOvbcmVAV3uYeydN+ZEyPbds2ebmL6SwD77x+XXfvL2qNmzK4ChzDCZyBC5fQgDtoggcE+vAKbzCxJta79WF9zkZXrHLnCObK+voFDQ6ekw==</latexit>

<latexit sha1_base64="g1Y4yQGEV9NzThZZTRXwn73mAS4=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSc3vVmlN3poWWgVuCGpTV7FV/umFCspgKTThWquM6qfZzLDUjnI7tbqZoiskQ92nHQIFjqvx8anWMTg0ToiiRpoVGU/bvRo5jpUZxYCZjrAdqUSvI/7ROpqMrP2cizTQVZHYoyjjSCSr+RiGTlGg+MgATyYxXRAZYYqJNOvbcmVAV3uYeydN+ZEyPbds2ebmL6SwD77x+XXfvL2qNmzK4ChzDCZyBC5fQgDtoggcE+vAKbzCxJta79WF9zkZXrHLnCObK+voFDQ6ekw==</latexit><latexit sha1_base64="g1Y4yQGEV9NzThZZTRXwn73mAS4=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSc3vVmlN3poWWgVuCGpTV7FV/umFCspgKTThWquM6qfZzLDUjnI7tbqZoiskQ92nHQIFjqvx8anWMTg0ToiiRpoVGU/bvRo5jpUZxYCZjrAdqUSvI/7ROpqMrP2cizTQVZHYoyjjSCSr+RiGTlGg+MgATyYxXRAZYYqJNOvbcmVAV3uYeydN+ZEyPbds2ebmL6SwD77x+XXfvL2qNmzK4ChzDCZyBC5fQgDtoggcE+vAKbzCxJta79WF9zkZXrHLnCObK+voFDQ6ekw==</latexit>

[image:2.595.324.507.139.422.2]

<latexit sha1_base64="g1Y4yQGEV9NzThZZTRXwn73mAS4=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDSc3vVmlN3poWWgVuCGpTV7FV/umFCspgKTThWquM6qfZzLDUjnI7tbqZoiskQ92nHQIFjqvx8anWMTg0ToiiRpoVGU/bvRo5jpUZxYCZjrAdqUSvI/7ROpqMrP2cizTQVZHYoyjjSCSr+RiGTlGg+MgATyYxXRAZYYqJNOvbcmVAV3uYeydN+ZEyPbds2ebmL6SwD77x+XXfvL2qNmzK4ChzDCZyBC5fQgDtoggcE+vAKbzCxJta79WF9zkZXrHLnCObK+voFDQ6ekw==</latexit> <latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit>

w2

<latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="UAucQw35IkZJXMJZS3jhGvwyt5k=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4Kkkv6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2tb2xubVd27N29/YPD6tHxk0oySahHEp7IToAV5UxQTzPNaSeVFMcBp+1gdFvo7WcqFUvEox6n1I/xQLCIEawN9fDSb/SrNafuzAqtArcENSir1a/+9MKEZDEVmnCsVNd1Uu3nWGpGOJ3YvUzRFJMRHtCugQLHVPn5zOoEnRsmRFEiTQuNZuzfjRzHSo3jwEzGWA/VslaQ/2ndTEdXfs5EmmkqyPxQlHGkE1T8jUImKdF8bAAmkhmviAyxxESbdOyFM6EqvC08kqeDyJie2LZt8nKX01kFXqN+XXfvnVrzpgyuAqdwBhfgwiU04Q5a4AGBAbzCG0ytqfVufVif89E1q9w5gYWyvn4BDXyekA==</latexit><latexit sha1_base64="5OklDCzJnyeNxwGEw0/tdoGXbdw=">AAACCXicbVDNSsNAGPxS/2qsWr16WSyCp5L0ot4ELx4rGltoQ9lsNunSzSbsbpQQ+ghe7dN4Eh/ChxHctD3Y1oEPhpldvvkmyDhT2nG+rdrW9s7uXn3fPmgcHh03TxrPKs0loR5JeSr7AVaUM0E9zTSn/UxSnASc9oLJXeX3XqhULBVPusion+BYsIgRrI30+DrqjJotp+3MgTaJuyQtWKI7av4Mw5TkCRWacKzUwHUy7ZdYakY4ndrDXNEMkwmO6cBQgROq/HIedYoujBKiKJVmhEZz9e+PEidKFUlgXiZYj9W6V4n/eYNcR9d+yUSWayrIYlGUc6RTVN2NQiYp0bwwBBPJTFZExlhiok079sqaUFXZVg4pszgyoae2bZu+3PV2NonXad+03QcH6nAG53AJLlzBLdxDFzwgEMMbvMPMmlkf1uei2Jq1bPgUVmB9/QI6kZ0V</latexit><latexit sha1_base64="5OklDCzJnyeNxwGEw0/tdoGXbdw=">AAACCXicbVDNSsNAGPxS/2qsWr16WSyCp5L0ot4ELx4rGltoQ9lsNunSzSbsbpQQ+ghe7dN4Eh/ChxHctD3Y1oEPhpldvvkmyDhT2nG+rdrW9s7uXn3fPmgcHh03TxrPKs0loR5JeSr7AVaUM0E9zTSn/UxSnASc9oLJXeX3XqhULBVPusion+BYsIgRrI30+DrqjJotp+3MgTaJuyQtWKI7av4Mw5TkCRWacKzUwHUy7ZdYakY4ndrDXNEMkwmO6cBQgROq/HIedYoujBKiKJVmhEZz9e+PEidKFUlgXiZYj9W6V4n/eYNcR9d+yUSWayrIYlGUc6RTVN2NQiYp0bwwBBPJTFZExlhiok079sqaUFXZVg4pszgyoae2bZu+3PV2NonXad+03QcH6nAG53AJLlzBLdxDFzwgEMMbvMPMmlkf1uei2Jq1bPgUVmB9/QI6kZ0V</latexit><latexit sha1_base64="pgt7e/aUqPh6UaONFIne6e+X0Ds=">AAACA3icbVBNS8NAFHypXzVWrWcvi0XwVBIv6k3w4rGCsYU2lM3mpV262YTdTaGE/gGv/TWexJ/hjxHctD3Y1oEHw8wub95EueDaeN63U9vbPzg8qh+7Jw339Oy82XjTWaEYBiwTmepFVKPgEgPDjcBerpCmkcBuNHmq/O4UleaZfDWzHMOUjiRPOKPGSp1hs+W1vSXILvHXpAVrDJs/gzhjRYrSMEG17vtebsKSKsOZwLk7KDTmlE3oCPuWSpqiDstlzDm5tkpMkkzZkYYs1b8/SppqPUsj+zKlZqy3vUr8z+sXJrkPSy7zwqBkq0VJIYjJSHUziblCZsTMEsoUt1kJG1NFmbHNuBtrYl1l2zikzEeJDT13XdfW5W+Xs0uC2/ZD23/xoA6XcAU34MMdPMIzdCAABjG8w8JZOB/O56rVmrOu9wI24Hz9AiipmwI=</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit><latexit sha1_base64="lEzryKbe+vId/yS+8EWOKAcaVGY=">AAACFHicbVBNS8NAEJ3Urxq/qh69BIvgqSRFUG9FLx4rGltoQ9lsNunSzSbsbpQS+hO82l/jSbx698cIbtocbOvAwOO9GebN81NGpbLtb6Oytr6xuVXdNnd29/YPaodHTzLJBCYuTlgiuj6ShFFOXEUVI91UEBT7jHT80W2hd56JkDThj2qcEi9GEachxUhp6uFl0BzU6nbDnpW1CpwS1KGs9qD20w8SnMWEK8yQlD3HTpWXI6EoZmRi9jNJUoRHKCI9DTmKifTymdWJdaaZwAoToZsra8b+3chRLOU49vVkjNRQLmsF+Z/Wy1R45eWUp5kiHM8PhRmzVGIVf1sBFQQrNtYAYUG1VwsPkUBY6XTMhTOBLLwtPJKnUahNT0zT1Hk5y+msArfZuG449xf11k0ZXBVO4BTOwYFLaMEdtMEFDBG8whtMjanxbnwYn/PRilHuHMNCGV+/DryelA==</latexit>

wl

<latexit sha1_base64="QJD2scsN7UlvwGUf7snxedOBr6E=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDS471qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CcCiezg==</latexit><latexit sha1_base64="QJD2scsN7UlvwGUf7snxedOBr6E=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDS471qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CcCiezg==</latexit><latexit sha1_base64="QJD2scsN7UlvwGUf7snxedOBr6E=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDS471qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CcCiezg==</latexit><latexit sha1_base64="QJD2scsN7UlvwGUf7snxedOBr6E=">AAACFHicbVBNS8NAEJ34WeNX1aOXxSJ4KokI6q3oxWNFYwttKJvNpl262YTdjVJCf4JX+2s8iVfv/hjBTZuDbR0YeLw3w7x5QcqZ0o7zba2srq1vbFa27O2d3b396sHhk0oySahHEp7IdoAV5UxQTzPNaTuVFMcBp61geFvorWcqFUvEox6l1I9xX7CIEawN9fDS471qzak700LLwC1BDcpq9qo/3TAhWUyFJhwr1XGdVPs5lpoRTsd2N1M0xWSI+7RjoMAxVX4+tTpGp4YJUZRI00KjKft3I8exUqM4MJMx1gO1qBXkf1on09GVnzORZpoKMjsUZRzpBBV/o5BJSjQfGYCJZMYrIgMsMdEmHXvuTKgKb3OP5Gk/MqbHtm2bvNzFdJaBd16/rrv3F7XGTRlcBY7hBM7AhUtowB00wQMCfXiFN5hYE+vd+rA+Z6MrVrlzBHNlff0CcCiezg==</latexit>

Figure 1: Plate representation of our proposed model.

Plates represent replication; E ×E is the number of

entity pairs in the dataset, S is the number of

sen-tences mentioning each entity pair andR is the

num-ber of relations. Arrows represent functions from input to output. Latent variables are represented as unshaded nodes. Factors over variables are represented as boxes.

2.2 Mention Representation

In the following section we review the Piecewise CNN (PCNN) architecture, first proposed by Zeng et. al. (2015), which is used as the basis for our feature representation.

Input Representation: A sentence,si consisting

ofl words is represented by two types of

embed-dings: word embeddings, Ei, and position

em-beddings,Pi relative to the entity pair. Following

Lin et. al. (2016), word embeddings were ini-tialized by running Word2Vec on the New York Times corpus and later fine-tuned; position em-beddings encode the position of the word relative

to KB entities, e1 ande2, mentioned in the

(3)

of embedding at each word position is equal to the word embedding dimension plus two times the po-sition embedding size (one popo-sition is encoded for each entity).

Convolution: Given an input sentence

represen-tation, we perform 1D convolution within a

win-dow of length l to extract local features.

As-sume we have df convolutional filters (F =

{f1, f2,· · ·, fdf}, fi ∈ R

l×d_{). The output of the}

i-th convolutional filter within thej-th window is:

cij=fi·wj−l+1:j+b (1≤j≤m+l−1)

Wherebis a bias term. We use zero padding when

the window slides out of the sentence boundaries.

Piecewise Max Pooling: The output of the

con-volutional layer ci is separated into three parts

(ci1, ci2, ci3) using the positions of the two

enti-ties in the sentence. Max pooling over time is then applied to each of these parts, followed by an el-ementwise tanh. The final sentence vector is de-fined as follows:

[x]ik= tanh(max

j (cikj)) (1≤i≤df,1≤k≤3)

2.3 Structured Minimally Supervised

Learning

Our proposed model is based on the PCNN repre-sentations described above, in addition to a latent variable model that reasons about missing data and ambiguous relations during learning and is

illus-trated in Figure1. The embedding for sentencei,

is used to define a factor over theith input sentence

and latent relation mention variablezi:

φPCNN(si, zi) =exi·θzi

wherexi is the representation for sentencesi, as

encoded by the piecewise CNN.

Another set of factors, φOR, link the

sentence-level mention variables,zi, to aggregate-level

vari-ablestj, representing whether relationj is

men-tioned betweene1ande2 in text. This is modeled

using a deterministic OR:

φOR(z, tj) =1¬tj⊕∃i:j=zi

where 1x is an indicator function that takes the

value 1 whenx is true. The choice of

determin-istic OR can be interpreted intuitively as follows:

if a proposition is true according totj, then it must

be extracted from at least one sentence in the train-ing corpus, on the other hand, if it is false, no sen-tences in the corpus can mention it.

Finally, we incorporate a set of factors that pe-nalize disagreement between observed relations in

the KB,dj, and latent variables tj, which

repre-sent whether relationjwas extracted from the text.

The penalties for disagreement with the KB are hyperparameters that are adjusted on held-out de-velopment data and incorporate entity frequency information from the KB, to model the intuition that more popular entities are less likely to have missing facts:

φA(tj, dj) =

  

 

e−αT_, _if_t

j= 0anddj= 1

e−αD_, _if_t

j= 1anddj= 0

1, otherwise

Putting everything together, the (unnormalized)

joint distribution over t, d andz conditioned on

sentences s mentioning a dyad is defined as

fol-lows:

P(d,t,z|s) ∝ |s|

Y

i=1

φPCNN(si, zi)×

|r|

Y

j=1

φOR(z, tj)φA(tj, dj)

µ

= exp(Sθ(s,z,t,d))

(1)

Here, µ is a tunable hyperparameter to adjust

impact of the disagreement penalty, and Sθ(·) is

the model score for a joint configuration of vari-ables, which corresponds to the log of the unnor-malized probability.

A standard conditional random field (CRF)

for-mulation would optimize model parameters,θ so

as to maximize marginal probability of the

ob-served KB relations, d conditioned on observed

sentences,s:

P(d|s) =X

z,t

P(d,t,z|s)

Computing gradients with respect toP(d|s)(and

marginalizing out z and t) is computationally

intractable, so instead we propose an approach that uses maximum-a-posteriori (MAP) parameter

learning (Taskar et al.,2004) and is inspired by the

latent structured SVM (Yu and Joachims,2009).

Given a large text corpus in which a set of

sen-tences,smention a specific pair of entities(e1, e2)

(4)

our goal is to minimize the structured hinge loss:

LSH(θ) =

max        0, max z∗ e,t∗e,d∗e

[Sθ(s,z∗e,t ∗ e,d

∗

e) +lHam(d∗e,d)] −max

z∗ g,t∗g

Sθ(s,z∗g,t∗g,d)

       (2)

Where lHam(d∗e,d) is the Hamming distance

be-tween the bit vector corresponding to the set of

observed relations holding between(e1, e2)in the

KB and those predicted by the model.

Minimiz-ingLSH(θ)can be understood intuitively as

adjust-ing the parameters so that configurations

consis-tent with observed relations in the KB,d, achieve

a higher model score than those with a large ham-ming distance from the observed configuration.

z∗_e corresponds to the most confusing

configura-tion of the sentence-level relaconfigura-tion menconfigura-tion vari-ables (i.e. one that has a large score and also a

large Hamming loss) and z∗_g corresponds to the

best configuration that is consistent with the ob-served relations in the KB.

This objective can be minimized using

stochas-tic subgradient descent. Fixingz∗_g andz∗_e to their

maximum values in Equation2, subgradients with

respect to the parameters can be computed as fol-lows:

∇θLSH(θ) =

  

 

0 ifLSH(θ)≤0,

∇θSθ(s,z∗e,t ∗ e,d

∗ e) −∇θSθ(s,z∗g,t

∗

g,d) otherwise

(3) =     

0 ifLSH(θ)≤0,

P

i∇θlogφPCNN(si,z∗e,i) −P

i∇θlogφPCNN(si,zg,i∗ ) otherwise

(4)

Because the second factor of the product in

Equation1does not depend onθ, it is

straightfor-ward to compute subgradients of the scoring

func-tion,∇Sθ(·), with fixed values ofz∗gandz∗eusing

backpropagation (Equation4).

Inference: The two inference problems,

corre-sponding to maximizing over hidden variables in

Equation 2 can be solved using a variety of

so-lutions; we experimented with A∗ search over

left-to-right assignments of the hidden variables. An admissible heuristic is used to upper-bound the maximum score of each partial hypothesis by maximizing over the unassigned PCNN factors, ignoring inconsistencies. This approach is guar-anteed to find an optimal solution, but can be slow and memory intensive for problems with many

variables. In preliminary experiments on devel-opment data, we found that local-search (Eisner

and Tromble, 2006) using both relation type and

mention search operators (Liang et al.,2010;

Rit-ter et al.,2013) usually finds an optimal solution

and also scales up to large training datasets; we use local search with 30 random restarts to com-pute argmax assignments for the hidden variables,

z∗_g andz∗_e, in all our experiments.

Bag-Size Adaptive Learning Rate: Since the

search space of the MAP inference problem in-creases exponentially as the number of hidden variables goes up, it becomes more difficult to find the exact argmax solution using local search, lead-ing to increased noise in the computed gradients. To mitigate the search-error problem in large bags of sentences, we dynamically modify the learning rate based on the number of sentences in each bag as follows:

λi=

  

 

λ, if|si|< β1

λ× β1

|s_i|, ifβ1≤ |si| ≤β2

λ×(β1 |si|)

2_,

otherwise

whereλiis the learning rate forith training entity

pair and β1/β2 are two tunable bag-size

thresh-olds. In Table 3 and Table 4, we see that this

strategy significantly improves performance, espe-cially when training on the larger NYTFB-280K dataset. We also experimented with this method for PCNN+ATT, but found that its performance did not improve.

3 Experiments

In Section 2, we presented an approach that com-bines the benefits of PCNN representations and structured learning with latent variables for

min-imally supervised relation extraction. In this

section we present the details of our evaluation methodology and experimental results.

Datasets: We evaluate our models on the

NYT-Freebase dataset (Riedel et al., 2010) which was

created by aligning relational facts from Freebase with the New York Times corpus, and has been used in a broad range of prior work on minimally supervised relation extraction. Several versions of this dataset have been used in prior work; to fa-cilitate the reproduction of prior results, we ex-periment with two versions of the dataset used by Riedel et. al. (2010) (henceforth NYTFB-68K) and Lin et. al. (2016) (NYTFB-280K). Statistics

of these datasets are presented in Table8. A more

(5)

Dataset NYTFB-68K NYTFB-280K

(Riedel et. al. 2010) (Lin et. al. 2016)

Entity pairs 67,946 280,275

Sentences 120,290 523,312

Table 1: Number of entity pairs and sentences in

the training portion of Riedel’s HELDOUT dataset

(NYTFB-68K) and Lin’s dataset (NYTFB-280K).

Window lengthl 3

Number of convolutional filtersdf 230

Word embedding dimensiondw 50

Position embedding dimensiondp 5

Batch sizeB 1

Table 2: Untuned hyperparameters in our experiments.

datasets used in prior work is also presented in Ap-pendix B.

Hyperparameters: Following Lin et. al. (2016),

we utilize word embeddings pre-trained on the NYT corpus using the word2vec tool, other pa-rameters are initialized using the method de-scribed by Glorot and Bengio (2010). The

Hoff-mann et. al. sentential evaluation dataset is

split into a development and test set and grid search on the development set was used to

de-termine optimal values for the learning rate λ

among {0.001,0.01}, KB disagreement penalty

scalar µ among{100,200,· · · ,2000} andβ1/β2

bag size threshold for the adaptive learning rate

among{10,15,· · ·,40}. Other hyperparameters

with fixed values are presented in Table2.

Neural Baselines: To demonstrate the

effective-ness of the our approach, we compare against

col-less universal schema (Verga et al., 2016) in

ad-dition to the PCNN+ATT model of Lin et. al. (2016). After training the Lin et. al. model to predict observed facts in the KB, we use its at-tention layer to make mention-level predictions as follows:

p(rj|xi) =

exp(rj·xi)

Pnr

k=1exp(rk·xi)

Whererjindicates the vector representation of the

jth relation.

Structured Baselines: In addition to

initializ-ing convolutional filters used in theφPCNN(·)

fac-tors randomly and performing structured

learn-ing of representations as in Equation 4, we also

experimented with variants of MultiR and DN-MAR, which are based on the structured

percep-tron (Collins, 2002), using fixed sentence

resentations: both traditional sparse feature

rep-resentations, in addition to pre-trained contin-uous representations generated using our best-performing reimplementation of PCNN+ATT. For the structured perceptron baselines, we also exper-imented with variants based on MIRA (Crammer

and Singer,2003), which we found to provide

con-sistent improvements. More details are provided in Appendix A.

3.1 Sentential Evaluation

In this work, we are primarily interested in mention-level relation extraction. For our first set

of experiments (Tables3and4), we use the

manu-ally annotated dataset created by (Hoffmann et al., 2011). Note that sentences in the Hoffman et. al. dataset were selected from the output of systems used in their evaluation, so it is possible there are high confidence predictions made by our systems that are not present. Therefore, we further validate our findings, by performing a manual inspection

of the highest confidence predictions in Table5.

NYTFB-68K Results: As illustrated in Table 3,

simply applying structured models (MultiR and DNMAR) with pre-trained sentence

representa-tions performs competitively. MIRA provides

consistent improvements for both sparse and dense

representations. PCNN+ATT outperforms most

latent-variable models on the sentential evalua-tion, we found this result to be surprising as the model was designed for extracting proposition-level facts. Col-less universal schema does not perform very well in this evaluation; this is likely due to the fact that it was developed for the KBP

slot filling evaluation (Ji et al., 2010), and only

uses the part of a sentence between two entities as an input representation, which can remove impor-tant context. Our proposed model, which jointly learns sentence representations using a structured latent-variable model that allows for the possiblity of missing data, achieves the best overall perfor-mance; its improvements over all baselines were found to be statistically significant according to a

paired bootstrap test (Efron and Tibshirani,1994;

Berg-Kirkpatrick et al.,2012).2

NYTFB-280K Results: When training on the

larger dataset provided by Lin et. al. (2016), lin-guistic features are not available, so only neural representations are included in our evaluation. As

illustrated in Table 4, PCNNNMAR also achieves

the best performance when training on the larger

[image:5.595.100.264.166.220.2]

(6)

Model Name DEV TEST

Fixed Sentence Representations

MultiR sparse (Hoffmann et al.,2011) 66.2 63.2

MultiR sparse MIRA 75.3 71.6

MultiR continuous 74.2 68.7

MultiR continuous MIRA 80.3 72.5

DNMAR sparse (Ritter et al.,2013) 77.9 70.1

DNMAR sparse MIRA 77.5 72.1

DNMAR continuous 80.2 70.0

DNMAR continuous MIRA 82.2 74.2

Jointly Learned Representations

PCNNNMAR 82.4 83.9

PCNNNMAR(bag size adaptive learning rate) 85.4 86.0

Baselines

col-less universal schema (Verga et al.,2016) 63.4 61.1

PCNN+ATT (Lin et al. (2016) code) 81.4 76.4

[image:6.595.113.487.61.205.2]

PCNN+ATT (our reimplementation with parameter tuning) 83.6 78.4

Table 3: AUC of sentential evaluation precision / recall curves for all models trained on NYTFB-68K.

Continu-ous sentence representation works as well as human-engineered sentence representation, and MIRA consistently

helps structured perceptron training. PCNN+ATT performs competitively while our PCNNNMAR (AdapLR) is

statistically significantly better (p-value of bootstrap is less than 0.05)

dataset; its improvements over the baselines are statistically significant. The AUC of most mod-els decreases on the Hoffmann et. al. senten-tial dataset when training on NYTFB-280K. This is not surprising, because the Hoffmann et. al. dataset is built by sampling sentences from pos-itive predictions of models trained on NYTFB-68K; changing the training data causes a differ-ence in the ranking of high-confiddiffer-ence predictions for each model, leading to the observed decline in performance against the Hoffmann et. al. dataset. To further validate our findings, we also manually inspect the models’ top predictions as described below.

Manual Evaluation: Because the Hoffmann et.

al. sentential dataset does not contain the high-est confidence predictions, we also manually in-spected each model’s top 500 predictions for the most frequent 4 relations, and report precision @ N to further validate our results. As shown in

Table5, for NYTFB-68K, PCNN+ATT performs

comparably on /location/contains3 and

/person/company, whereas our model has a considerable advantage on the other two relations.

For NYTFB-280K, our model performs

consis-tently better on all four relations compared with PCNN+ATT. When training on the larger

NYTFB-280K dataset, we observe trend of increasing

mention-level P@N for PCNNNMAR, however the performance of PCNN+ATT appears to decrease. We investigate this phenomenon further below.

Performance at Extracting New Facts: To

ex-plain PCNN+ATT’s drop in mention-level

perfor-3_{/location/contains}_{is the most frequent relation} in the Hoffmann et. al. dataset.

mance after training on the larger NYTFB-280K dataset, our hypothesis is that the larger KB-supervised dataset not only contains more true positive training examples but also more false neg-ative examples. This biases models toward pre-dicting facts about popular entities, which are likely to exist in Freebase. To provide evidence in support of this hypothesis, we divide the manu-ally annotated dataset from Hoffmann et. al. into two categories: mentions of facts found in Free-base, and those that are not; this distribution is

presented in the Table 6. In Table 7, we present

a breakdown of model performance on these two subsets. For PCNN+ATT, although the AUC of in-Freebase mentions on the test set increases af-ter training on the larger NYTFB-280K, its Out-Of-Freebase AUC on both dev and test sets drops significantly, which clearly illustrates the problem of increasing false negatives during training. In contrast, our model, which explicitly allows for the possibility of missing data in the KB during learning, has relatively stable performance on the two types of mentions, as the amount of weakly-supervised training data is increased.

3.2 Held-Out Evaluation

In Section 3.1, we evaluated the results of