Mitigating Metric Bias in Minimum Bayes Risk Decoding

Autor: Kovacs, Geza, Deutsch, Daniel, Freitag, Markus

While Minimum Bayes Risk (MBR) decoding using metrics such as COMET or MetricX has outperformed traditional decoding methods such as greedy or beam search, it introduces a challenge we refer to as metric bias. As MBR decoding aims to produce translat

Externí odkaz: http://arxiv.org/abs/2411.03524

GPT-4o System Card

Autor: OpenAI, Hurst, Aaron, Lerer, Adam, Goucher, Adam P., Perelman, Adam, Ramesh, Aditya, Clark, Aidan, Ostrow, AJ, Welihinda, Akila, Hayes, Alan, Radford, Alec, Mądry, Aleksander, Baker-Whitcomb, Alex, Beutel, Alex, Borzunov, Alex, Carney, Alex, Chow, Alex, Kirillov, Alex, Nichol, Alex, Paino, Alex, Renzin, Alex, Passos, Alex Tachard, Kirillov, Alexander, Christakis, Alexi, Conneau, Alexis, Kamali, Ali, Jabri, Allan, Moyer, Allison, Tam, Allison, Crookes, Amadou, Tootoochian, Amin, Tootoonchian, Amin, Kumar, Ananya, Vallone, Andrea, Karpathy, Andrej, Braunstein, Andrew, Cann, Andrew, Codispoti, Andrew, Galu, Andrew, Kondrich, Andrew, Tulloch, Andrew, Mishchenko, Andrey, Baek, Angela, Jiang, Angela, Pelisse, Antoine, Woodford, Antonia, Gosalia, Anuj, Dhar, Arka, Pantuliano, Ashley, Nayak, Avi, Oliver, Avital, Zoph, Barret, Ghorbani, Behrooz, Leimberger, Ben, Rossen, Ben, Sokolowsky, Ben, Wang, Ben, Zweig, Benjamin, Hoover, Beth, Samic, Blake, McGrew, Bob, Spero, Bobby, Giertler, Bogo, Cheng, Bowen, Lightcap, Brad, Walkin, Brandon, Quinn, Brendan, Guarraci, Brian, Hsu, Brian, Kellogg, Bright, Eastman, Brydon, Lugaresi, Camillo, Wainwright, Carroll, Bassin, Cary, Hudson, Cary, Chu, Casey, Nelson, Chad, Li, Chak, Shern, Chan Jun, Conger, Channing, Barette, Charlotte, Voss, Chelsea, Ding, Chen, Lu, Cheng, Zhang, Chong, Beaumont, Chris, Hallacy, Chris, Koch, Chris, Gibson, Christian, Kim, Christina, Choi, Christine, McLeavey, Christine, Hesse, Christopher, Fischer, Claudia, Winter, Clemens, Czarnecki, Coley, Jarvis, Colin, Wei, Colin, Koumouzelis, Constantin, Sherburn, Dane, Kappler, Daniel, Levin, Daniel, Levy, Daniel, Carr, David, Farhi, David, Mely, David, Robinson, David, Sasaki, David, Jin, Denny, Valladares, Dev, Tsipras, Dimitris, Li, Doug, Nguyen, Duc Phong, Findlay, Duncan, Oiwoh, Edede, Wong, Edmund, Asdar, Ehsan, Proehl, Elizabeth, Yang, Elizabeth, Antonow, Eric, Kramer, Eric, Peterson, Eric, Sigler, Eric, Wallace, Eric, Brevdo, Eugene, Mays, Evan, Khorasani, Farzad, Such, Felipe Petroski, Raso, Filippo, Zhang, Francis, von Lohmann, Fred, Sulit, Freddie, Goh, Gabriel, Oden, Gene, Salmon, Geoff, Starace, Giulio, Brockman, Greg, Salman, Hadi, Bao, Haiming, Hu, Haitang, Wong, Hannah, Wang, Haoyu, Schmidt, Heather, Whitney, Heather, Jun, Heewoo, Kirchner, Hendrik, Pinto, Henrique Ponde de Oliveira, Ren, Hongyu, Chang, Huiwen, Chung, Hyung Won, Kivlichan, Ian, O'Connell, Ian, Osband, Ian, Silber, Ian, Sohl, Ian, Okuyucu, Ibrahim, Lan, Ikai, Kostrikov, Ilya, Sutskever, Ilya, Kanitscheider, Ingmar, Gulrajani, Ishaan, Coxon, Jacob, Menick, Jacob, Pachocki, Jakub, Aung, James, Betker, James, Crooks, James, Lennon, James, Kiros, Jamie, Leike, Jan, Park, Jane, Kwon, Jason, Phang, Jason, Teplitz, Jason, Wei, Jason, Wolfe, Jason, Chen, Jay, Harris, Jeff, Varavva, Jenia, Lee, Jessica Gan, Shieh, Jessica, Lin, Ji, Yu, Jiahui, Weng, Jiayi, Tang, Jie, Yu, Jieqi, Jang, Joanne, Candela, Joaquin Quinonero, Beutler, Joe, Landers, Joe, Parish, Joel, Heidecke, Johannes, Schulman, John, Lachman, Jonathan, McKay, Jonathan, Uesato, Jonathan, Ward, Jonathan, Kim, Jong Wook, Huizinga, Joost, Sitkin, Jordan, Kraaijeveld, Jos, Gross, Josh, Kaplan, Josh, Snyder, Josh, Achiam, Joshua, Jiao, Joy, Lee, Joyce, Zhuang, Juntang, Harriman, Justyn, Fricke, Kai, Hayashi, Kai, Singhal, Karan, Shi, Katy, Karthik, Kavin, Wood, Kayla, Rimbach, Kendra, Hsu, Kenny, Nguyen, Kenny, Gu-Lemberg, Keren, Button, Kevin, Liu, Kevin, Howe, Kiel, Muthukumar, Krithika, Luther, Kyle, Ahmad, Lama, Kai, Larry, Itow, Lauren, Workman, Lauren, Pathak, Leher, Chen, Leo, Jing, Li, Guy, Lia, Fedus, Liam, Zhou, Liang, Mamitsuka, Lien, Weng, Lilian, McCallum, Lindsay, Held, Lindsey, Ouyang, Long, Feuvrier, Louis, Zhang, Lu, Kondraciuk, Lukas, Kaiser, Lukasz, Hewitt, Luke, Metz, Luke, Doshi, Lyric, Aflak, Mada, Simens, Maddie, Boyd, Madelaine, Thompson, Madeleine, Dukhan, Marat, Chen, Mark, Gray, Mark, Hudnall, Mark, Zhang, Marvin, Aljubeh, Marwan, Litwin, Mateusz, Zeng, Matthew, Johnson, Max, Shetty, Maya, Gupta, Mayank, Shah, Meghan, Yatbaz, Mehmet, Yang, Meng Jia, Zhong, Mengchao, Glaese, Mia, Chen, Mianna, Janner, Michael, Lampe, Michael, Petrov, Michael, Wu, Michael, Wang, Michele, Fradin, Michelle, Pokrass, Michelle, Castro, Miguel, de Castro, Miguel Oom Temudo, Pavlov, Mikhail, Brundage, Miles, Wang, Miles, Khan, Minal, Murati, Mira, Bavarian, Mo, Lin, Molly, Yesildal, Murat, Soto, Nacho, Gimelshein, Natalia, Cone, Natalie, Staudacher, Natalie, Summers, Natalie, LaFontaine, Natan, Chowdhury, Neil, Ryder, Nick, Stathas, Nick, Turley, Nick, Tezak, Nik, Felix, Niko, Kudige, Nithanth, Keskar, Nitish, Deutsch, Noah, Bundick, Noel, Puckett, Nora, Nachum, Ofir, Okelola, Ola, Boiko, Oleg, Murk, Oleg, Jaffe, Oliver, Watkins, Olivia, Godement, Olivier, Campbell-Moore, Owen, Chao, Patrick, McMillan, Paul, Belov, Pavel, Su, Peng, Bak, Peter, Bakkum, Peter, Deng, Peter, Dolan, Peter, Hoeschele, Peter, Welinder, Peter, Tillet, Phil, Pronin, Philip, Tillet, Philippe, Dhariwal, Prafulla, Yuan, Qiming, Dias, Rachel, Lim, Rachel, Arora, Rahul, Troll, Rajan, Lin, Randall, Lopes, Rapha Gontijo, Puri, Raul, Miyara, Reah, Leike, Reimar, Gaubert, Renaud, Zamani, Reza, Wang, Ricky, Donnelly, Rob, Honsby, Rob, Smith, Rocky, Sahai, Rohan, Ramchandani, Rohit, Huet, Romain, Carmichael, Rory, Zellers, Rowan, Chen, Roy, Chen, Ruby, Nigmatullin, Ruslan, Cheu, Ryan, Jain, Saachi, Altman, Sam, Schoenholz, Sam, Toizer, Sam, Miserendino, Samuel, Agarwal, Sandhini, Culver, Sara, Ethersmith, Scott, Gray, Scott, Grove, Sean, Metzger, Sean, Hermani, Shamez, Jain, Shantanu, Zhao, Shengjia, Wu, Sherwin, Jomoto, Shino, Wu, Shirong, Shuaiqi, Xia, Phene, Sonia, Papay, Spencer, Narayanan, Srinivas, Coffey, Steve, Lee, Steve, Hall, Stewart, Balaji, Suchir, Broda, Tal, Stramer, Tal, Xu, Tao, Gogineni, Tarun, Christianson, Taya, Sanders, Ted, Patwardhan, Tejal, Cunninghman, Thomas, Degry, Thomas, Dimson, Thomas, Raoux, Thomas, Shadwell, Thomas, Zheng, Tianhao, Underwood, Todd, Markov, Todor, Sherbakov, Toki, Rubin, Tom, Stasi, Tom, Kaftan, Tomer, Heywood, Tristan, Peterson, Troy, Walters, Tyce, Eloundou, Tyna, Qi, Valerie, Moeller, Veit, Monaco, Vinnie, Kuo, Vishal, Fomenko, Vlad, Chang, Wayne, Zheng, Weiyi, Zhou, Wenda, Manassra, Wesam, Sheu, Will, Zaremba, Wojciech, Patil, Yash, Qian, Yilei, Kim, Yongjik, Cheng, Youlong, Zhang, Yu, He, Yuchen, Zhang, Yuchen, Jin, Yujia, Dai, Yunxing, Malkov, Yury

GPT-4o is an autoregressive omni model that accepts as input any combination of text, audio, image, and video, and generates any combination of text, audio, and image outputs. It's trained end-to-end across text, vision, and audio, meaning all inputs

Externí odkaz: http://arxiv.org/abs/2410.21276

Beyond Human-Only: Evaluating Human-Machine Collaboration for Collecting High-Quality Translation Data

Autor: Liu, Zhongtao, Riley, Parker, Deutsch, Daniel, Lui, Alison, Niu, Mengmeng, Shah, Apu, Freitag, Markus

Collecting high-quality translations is crucial for the development and evaluation of machine translation systems. However, traditional human-only approaches are costly and slow. This study presents a comprehensive investigation of 11 approaches for

Externí odkaz: http://arxiv.org/abs/2410.11056

Improving Statistical Significance in Human Evaluation of Automatic Metrics via Soft Pairwise Accuracy

Autor: Thompson, Brian, Mathur, Nitika, Deutsch, Daniel, Khayrallah, Huda

Selecting an automatic metric that best emulates human annotators is often non-trivial, because there is no clear definition of "best emulates." A meta-metric is required to compare the human judgments to the automatic metric scores, and metric ranki

Externí odkaz: http://arxiv.org/abs/2409.09598

MS-IMAP -- A Multi-Scale Graph Embedding Approach for Interpretable Manifold Learning

Autor: Deutsch, Shay, Yelibi, Lionel, Lin, Alex Tong, Kannan, Arjun Ravi

Deriving meaningful representations from complex, high-dimensional data in unsupervised settings is crucial across diverse machine learning applications. This paper introduces a framework for multi-scale graph network embedding based on spectral grap

Externí odkaz: http://arxiv.org/abs/2406.02778