A quick guide to Amazon’s 50-Plus papers on Emnlp 2024

Large Language Models (LLMs) have come to dominate the area of ​​natural-language processing, so it is not surprising that they also dominate the research that Amazon scientists present at this year’s conference on empirical methods in natural-language procedure (Emnlp ). LLM training is the subject with the largest number of Amazon papers, which are closely followed by strategies to mitigate incorrect information in LLMS ‘output -including but not limited to hallucinations. At the same time, a number of papers use LLMs for topics of traditional interest in Amazon, such as speech, recommendation system and information about information. (Papers marked with stars were accepted Results of Emnlp.)

AI agents

Marco: Multi-agent real-time chat orchestration
Anubhav Shrimal, Shervin Malmasi, Criti Biswas, Swarnalatha Raghuraman, Anish Nediyanchath, Yi Zhang, Promod Yenigalla

Generation code

CODEFORT: Robust training for code generation models
Yuhao Zhang, Shiqi Wang, Haifeng Qian, Zijian Wang, Mingyue Shang, Linbo Liu, Sanjay Krishna Gouda, Baishakhi Ray, Murali Krishna Ramanathan, Xiaofei Ma, Anoo Deoras

Socratic Human Feedback (Sohf): Management Strategies Expert for LLM coding generation
Subramanian Chidambaram, Erran Li, My Bai, Xiaopng Li, Kaixiang Lin, Xiong Zhou, Alex C. Williams

Structural Object Language Modeling (SOLM): Native Structured Objects Generation in accordance with complex schedules with self -monitored denoising
Amir Tavanaei, Kee Kiat Koo, Hareddin Ceker, Shaobai Jiang, Qi Li, Julien Han, Karim Bouyarmane

Contrastive decoding

Explanation and improvement of contrastive decoding by extrapolating the likelihood of a huge and hypothetical LM
Haw-Shiuan Chang, Nanyun Money, Mohit Bansal, Anil Ramakrishna, Tagyoung Chung

Data integration

Astra: Automatic Schedule -Matching Machine Translation
Tarang Chugh, Deepak Zambre

Learning Natural Language Explanations at Generalizable Device Matching
Somin Wadhwa, Adit Krishnan, Runhui Wang, Byron C. Wallace, Chris (Luyang) King

Pretraining and Finetuning Language Models on Geospatial Network to accurate addressing Matching
SAKET MAHESHWARY, ARPAN PAUL, SAURABH SOHONEY

Download Increased Spelling Correction for E-Commerce Applications
Xuan Guo, Rohit Patki, Dante Everaert, Christopher Potts

Dataset Distillation

Text Data Set Distillation via Language Model Development
Yefan Tao, Chris (Luyang) Kong, Andrey can, Laurent Callot

Dallme Framework offers in “Text Data Set Distillation via Language Model Embedding” begins by using a language model to transform raw text data into embedded vectors. A set of distilled vectors is then derived in the embedding space through a process designed to encapsulate maximum information. Finlly translates the VEC2TEXT model these distilled vectors back to text form.

Understanding Document

DOCKD: Knowledge actillation from LLMS to Open World Document Understanding Models
Sungnyun Kim, Haofu Liao, Srikar Appalaraju, Money Tang, Zhuowen Tu, Pleased Kumar Satzoda, R. Manmatha, Vijay Mahadevan, Stefano Soatto

Information collection

Evaluation of D-Merit of Partial-Annotation on Information on Information
Royi Rassin, Yaron Fairstein, Oren Kalinsky, Guy Kushilevitz, Nachshon Cohen, Alexander Libov, Yoav Goldberg

Identification of high consideration of e-commerce-seeking queries
Zhiyu Chen, Jason Choi, Besnik Fetahu, Shervin Malmasi

Learning when to pick up what to rewrite and how to react in conversation QA*
Nirmal Roy, Leonardo Ribeiro, Rexhina Blloshmi, Kevin Small

Natural-language understanding

Intent Detection In The Age of LLMS
Gaurav Arora, Shreya Jain, Srujana Merugu

Prediction of Entity Salience in extremely short documents
Ben Bullough, Harrison Lundberg, Chen Hu, Weihang Xiao

LLM evaluation

Axcel: Explainable Automated consists of evaluation using LLMs*
P aditya Sreekar, Sahil Verma, Suransh Chopra, Sarik Ghazarian, Abhishek Persad, Narayanan Sadagopan

Exact Model Benchmarking with only a few observations
Riccardo Fogliato, Pratik Patil, Nil-Jana Akpinar, Mathew Monfort

Llm fine tuning

Adazeta: Adaptive Zero-Order Tensor-Togadapption to Memory Effective Large Language Models Finely Ticket
Yifan Yang, Kai Zhen, Ershad Banijamali, Thanasis Mouchtaris, Zheng Zhang

ROSELORA: Row and pillar-wise sparse adaptation of low rank of pre-formed language model for knowledge editing and fine tuning
HAOYU WANG, TANCI LIU, RUirui Li, Monica Cheng, Tuo Zhao, Jing Gao

Llms to speech

Talworthy Instruction -set language models
Hyundong Cho, Nicolaas Jedema, Leonardo Ribeiro, Karishma Sharma, Pedro Szekely, Alessandro Moschitti, Ruben Janssen, Jonathan May

LLM Incorrect information restriction

ECON: On detection and solution of conflicts
Cheng Jiayang, Chunkit Chan, Qianqian Zhuang, Lin Qiu, Tianhang Zhang, Tengxiao Liu, Yangqiu Song, Yue Zhang, Pengfei Liu, Zheng Zhang

Generative Undergraf Fund for Knowledge Graph – Lordnet Dialogue Generation
Jinyoung Park, Minseok Joo, Joo-Kyung Kim, Hyunwoo J. Kim

Hallumeave: Fink -grained Hallucination flooding using reasoning chain
Shayan Ali Akbar, MD Mosharaf Hossain, Tess Wood, Si-Chi Chin, Erica Salinas, Victor Alvarez, Erwin Cornejo

Nowledge-centered hallucination detection
Xiangkun Hu, Dongyu Ru, Lin Qiu, Qipeng Guo, Tianhang Zhang, Yang Xu, Yun Luo, Pengfei Liu, Zheng Zhang, Yue Zhang

LLM Reasoning

Auto-Volve: Improving the great language model’s performance through self-razing of frames*
Krishna Aswani, Alex Lu, Pranav Patankar, Priya Dhalwani, Iris Tan, Jayant Ganeshmohan, Simon Lacasse

Llm self -correction

LLM Self -correction with Criminal: Definion, criticism and refinement for improved after instructions with multiple restrictions
Thomas Palmeira Ferraz, Kartik Mehta, Yu-Hsiang Lin, Haw-Shiuan Chang, Shereen Oraby, Sijia Liu, Vivek Subramanian, Tagyoung Chung, Mohit Bansal, Nanyun Money

LLM training

Dance in chains: Reconciling instruction by and faithfulness in language models
Zhengxuan wu, yuhao zhang, money qi, yumo xu, rujun male, yian zhang, jifan chen, bonan min, zhiheng huang

Them: Distribution edited model for training with mixed data distribution
Dhananjay Ram, Aditya Rawal, Momchil Hardalov, Nikolaos Pappas, Sheng Zha

Distribution edited model D.) Described in “Them: Distribution Edited Model for Workout with Mixed Data Distributions” Results from fine tuning a prior model (Θ) we n Individual data distributions (D.I) And combine the resulting models with basic element-vis vector operations. Here are the extracted distribution vectors (∆∆Di ) Multiplied by weight coefficients and the weighted sum is added to the base model.

Evolutionary contrastive distillation to line model adjustment
Julian Katz-Samuelles, Zheng Li, Hyokun Yun, Priyanka Nigam, Yi Xu, Vaclav Petrick, Bing Yin, Trishul Chilimbi

Jump, jump, jump to convergence: dynamics in learning frequency transitions for improved training of large language models
Shreyas Subramanian, Vignnesh Ganapathiraman, Corey Barrett

Learning Reception
Magdalena Kaiser, Patrick Ernst, Gyuri Szarvas

Quality Questions: Evaluation of Synthetic Data for Tool -Using LLMs
Shadi Iskander, Nachshon Cohen, Zohar Karnin, Ori Shapira, Sofia Tolmach

Inquiry yourself

Amazonqac: A large -scale, naturalistic query AutocompleTe data set
Dante Everaer, Rohit Patki, Tianqi Zheng, Christopher Potts

Dial: Diversity AWARE LISTWISE RANKING FOR THE QUERY AUTO COMPLETE
Sonali Singh, Sachin Farfade, Prakash Mandayam Comar

Answering questions

RAG-QA Arena: Evaluation of Domain Robustness for Long-Fallened-Fetch-Augmented Questions Answer
Rujun he, yuhao zhang, money qi, yumo xu, jenyu wang, lan liu, william yang wang, bonan min, vittorio castelli

Retrieving contextual information to questions about long form answer using weak supervision
Philipp Christmann, Svitlana Vakulenko, Ionut Teodor Sorodoc, Bill Byrne, Adrià de Gispert

Recumend systems

Effective Point-PARWISE learning-to-rang to news recommendation
Nithish Kannen SentHilkumar, Yao Ma, Gerrit van den Burg, Jean Baptiste Faddoul

An illustration of Glimpese Framework offers in the “Effict PointWise-Pairwise Learning-to-Rank for news recommendation”. Glimpses adopt a multitask approach in which a pre-induced language model is fine-tuned on both the reduction prediction task and the paired Préer task. Under the inferency, the raised predictions are used to produce an initial point of location, which is then improved with one or more right-left (RTL) passage paired comparisons.

Pearl: Preference extraction with copy increase and fetch with LLM agents
Vijit Malik, Akshay Jagatap, Vinayak Puranik, Anirban Majumder

Sequential llm frame for fashion recommendation
He liu, Xianfeng Tang, Tianlang Chen, Jiapeng Liu, Indu Indu, Henry Peng Zou, Peng Dai, Roberto Fernandez Galan, Mike Porter, Dongmei Jia, Ning Zhang, Lian Xiong

AI Responsible

Attribute -controlled fine tuning for large language models: A case study on detoxification
Tao Meng, Ninareh Mehrabi, Palash Goyal, Anil Ramakrishna, Aram Galstyan, Richard Zemel, Kai-Wei Chang, Rahul Gupta, Charith Peris

Flirt: FEEDBACK LOOP IN-CONTEXT RED TEAMING
Ninareh Mehrabi, Palash Goyal, Christophe Dupuy, Qian Hu, Shalini Ghosh, Richard Zemel, Kai-Wei Chang, Aram Galstyan, Rahul Gupta

The size of the order for LLM -Membership Inferencies
Ronging Zhang, Martin Bertran Lopez, Aaron Roth

Synthetic Data Reration

Corrsynth: A correlated sampling method for different data set generation from LLMS
Suhas Kowshik, Abhishek Divekar, Vijit Malik

Data Counselor: Dynamic Data Courage for Security Adjustment of Large Language Models
Fei Wang, Ninareh Mehrabi, Palash Goyal, Rahul Gupta, Kai-Wei Chang, Aram Galstyan

Evaluation of Differentially Private Synthetic Data Reration In Domains With High Action
Krithika Ramesh, Nupoor Gandhi, Pulkit Madaan, Lisa Bauer, Charith Peris, Anjalie Field

Synthesizrr: Generation of Different Data Oetts with Aggroea Collection
Abhishek Divekar, Greg Durrett

Abstract depiction of the procedure suggested in “Synthesizr: Generation of different data sets with increase in retrieval”. The contentsourcing stage picks up K Unique documents {r1…, R.K} from a large corpus for each covariat in context xIcl. The assignment invitation stage uses a parameterized context -fining prompt, S.τthat take parameters R.Invis (Instruction Invention), R.k (A retrieved document) and V (yIcl) (The verbalized measurement label). A generalist teacher LLM generates authentication in synthetic covariat. Each example of context thus produces K Unique synthetic example {X̃1…, x̃K}As we include in the data set with target YIcl.

Text classification

Distance -conscious calibration to pre -formed language models*
Alberto Gasparin, Gianluca Detommaso

PERFORMANCE GUIDE LLM-VIDEMENTIFTING TO EFFECTIVE TEXT CLASSIFICATION IN SCALE

Flavio di Palo, Prateek Singhi, Bilal Fadlallah

Fast-displaced Muti-Task Taxonomic Transformation (PTMTTAX forms)
Rajashkar Vasantha, Nhan Nguyen, Yue Zhang

Summary text

Prominent information that asks to control content in quick-based abstract summary
Lei Xu, Asad Karim, Sketshot Dingliwal, Aparna Elanganvan

Leave a Comment