The DOI matches its Crossref record; 57 of the 57 references it cites were found in a registry or an index; 4 of the 9 sources its claims cite were checked for support.
Sequence-Level Knowledge Distillation
Open accessAbstract
Neural machine translation (NMT) offers a novel alternative formulation of translation that is potentially simpler than statistical approaches.However to reach competitive performance, NMT models need to be exceedingly large.In this paper we consider applying knowledge distillation approaches (Bucila et al., 2006;Hinton et al., 2015) that have proven successful for reducing the size of neural models in other domains to the problem of NMT.We demonstrate that standard knowledge distillation applied to word-level prediction can be effective for NMT, and also introduce two novel sequence-level versions of knowledge distillation that further improve performance, and somewhat surprisingly, seem to eliminate the need for beam search (even when applied on the original teacher model).Our best student model runs 10 times faster than its state-of-the-art teacher with little loss in performance.It is also significantly better than a baseline model trained without knowledge distillation: by 4.2/1.7 BLEU with greedy decoding/beam search.Applying weight pruning on top of knowledge distillation results in a student model that has 13× fewer parameters than the original teacher model, with a decrease of 0.4 BLEU.
Overview
AI snapshot
Written from the paper’s full text.
TL;DR
The paper applies knowledge distillation to neural machine translation, showing word-level distillation helps and introducing sequence-level variants (Seq-KD and Seq-Inter) that train a smaller student on teacher beam-search outputs. The best student runs 10× faster than its teacher with little BLEU loss, and pruning on top yields 13× fewer parameters with a 0.4 BLEU drop.
Key findings
- Word-level knowledge distillation improves over the baseline, but sequence-level knowledge distillation does better on English→German and similarly on Thai→English; combining them gives further gains for the 2×300 and 2×100 models, though not for the 2×500 model. For English→German 2×500, baseline BLEU is 14.7/17.6 (greedy/beam), Seq-KD is 18.9/19.0, and Seq-KD + Seq-Inter is 18.9/19.3, i.e. +4.2/+1.7 BLEU over baseline.
- Sequence-level distillation reduces the need for beam search: greedy decoding with a fine-tuned teacher (Baseline + Seq-Inter) reaches 19.6 BLEU versus 19.5 for beam search with the original teacher. Seq-KD models also assign much more mass to the approximate mode: 16.9% for the 2×500 English→German Seq-KD model versus 0.9% for the baseline.
- The student model can run 10× faster with greedy decoding than the teacher with beam search on GPU: 1051.3 vs 101.9 source words/sec.
- Weight pruning on top of distillation is orthogonal: pruning 80% of weights in the 2×500 Seq-KD + Seq-Inter student gives 13× fewer parameters than the original teacher with only 0.4 BLEU decrease; pruning 90% gives 8m parameters (26× fewer) with a 1.0 BLEU decrease.
Methods
- Uses an attentional encoder-decoder NMT architecture from Luong et al. (2015). Two settings: English→German WMT 2014 (4m sentences, top 50k vocabulary, teacher 4×1000 LSTM, students 2×300 and 2×500) and Thai→English IWSLT 2015 (90k sentences, 25k vocabulary, teacher 2×500, student 2×100). Evaluation is tokenized BLEU with multi-bleu.perl.
- Three distillation variants: Word-KD (cross-entropy with teacher word distribution plus NLL; α tested in {0.5, 0.9}, with 0.5 better); Seq-KD (student trained on teacher beam-search output, beam K=5); and Seq-Inter (student trained on the teacher beam hypothesis with highest smoothed sentence BLEU to gold, beam K=35, fine-tuned from a pretrained model with learning rate 0.1; English→German Seq-Inter data generated on ~50% of the training set). Methods can be combined.
- Weight pruning removes weights with lowest absolute values, then retrains on Seq-KD data with learning rate 0.2 and fine-tunes toward Seq-Inter data with learning rate 0.1. Speed is measured on GPU (GeForce GTX Titan X), CPU, and smartphone (Samsung Galaxy 6).
Limitations
- Stated: the mode approximation for sequence-level distillation is poor because it approximates an exponentially sized distribution with a single sample; Seq-Inter’s two-sequence interpolation is nonideal because it doubles training data and requires training on both teacher-generated and true sequences conditioned on the same source, which are often different. The paper also calls its illustrative noise function “simple (unrealistic).”
- Stated: experiments cover only English→German and Thai→English NMT; the 4×1000 model could not be opened on the smartphone; larger beam sizes than K=5 did not improve results; and English→German Seq-Inter data was generated on only ~50% of the training set for efficiency.
- Evident from the text: no statistical significance tests are reported, hyperparameter searches are narrow (τ ∈ {1, 1.5, 2}; α ∈ {0.5, 0.9}), and compression is compared mainly against pruning rather than other compression families.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
26 Own result · 13 Own interpretation · 9 No basis given · 3 Sources not found · 2 Partly supported · 1 Supported · 1 Supported so far · accuracy not yet measured
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“While both simple and surprisingly accurate, NMT systems typically need to have very high capacity in order to perform well: Sutskever et al. (2014) used a 4-layer LSTM with 1000 hidden units per layer (herein 4×1000) and Zhou et al. (2016) obtained state-of-the-art results on English → French with a 16-layer LSTM with 512 units per layer.”
§ Sequence-Level Knowledge Distillation > 1 Introduction
What the sources say
- Not checked — no full text of the sourcecites Sutskever et al. (2014) source
- Supportedcites Zhou et al. (2016) source
The Model settings section states that every LSTM layer has 512 memory cells. The Introduction and Conclusion report a depth-16 model that achieved state-of-the-art results on WMT’14 English-to-French, supporting the specific example attributed to Zhou et al. (LLM, conf=0.93)
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Our best student model runs 10 times faster than its state-of-the-art teacher with little loss in performance.”
§ Sequence-Level Knowledge Distillation > Abstract
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Sequence-level interpolation (Seq-Inter), in addition to improving models trained via Word-KD and Seq-KD, also improves upon the original teacher model that was trained on the actual data but finetuned towards Seq-Inter data (Baseline + Seq-Inter).”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In fact, greedy decoding with this fine-tuned model has similar performance (19.6) as beam search with the original model (19.5), allowing for faster decoding even with an identically-sized model.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Our results suggest that this is indeed the case: the probability mass that Seq-KD models assign to the approximate mode is much higher than is the case for baseline models trained on original data (Table 1: p(t = ŷ)).”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
hedgedcorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Sequence-level interpolation (Seq-Inter), in addition to improving models trained via Word-KD and Seq-KD, also improves upon the original teacher model that was trained on the actual data but finetuned towards Seq-Inter data (Baseline + Seq-Inter).”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that while word-level knowledge distillation (Word-KD) does improve upon the baseline, sequence-level knowledge distillation (Seq-KD) does better on English → German and performs similarly on Thai → English.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that while word-level knowledge distillation (Word-KD) does improve upon the baseline, sequence-level knowledge distillation (Seq-KD) does better on English → German and performs similarly on Thai → English.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Seq-Inter offers a compromise between the two, with the greedily-decoded sequence accounting for 7.6% of the distribution.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that while word-level knowledge distillation (Word-KD) does improve upon the baseline, sequence-level knowledge distillation (Seq-KD) does better on English → German and performs similarly on Thai → English.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- It may state two separate claims as one. 15
- Probably not a substantial claim. 6
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Researchers have particularly noted that large networks seem to be necessary for training, but learn redundant representations in the process (Denil et al., 2013).”
§ Sequence-Level Knowledge Distillation > 1 Introduction
What the sources say
- Partly supportedcites Denil et al., 2013 sourceExperimental — accuracy not yet measured
The Abstract (p. 1) reports significant redundancy in model parameters, and the Introduction (p. 2) explains that individual feature weights can be redundant. The paper does not say that large networks are necessary for training; its finding concerns redundant parameterization, not necessarily redundant representations. (LLM, conf=0.85)
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“In this paper we consider applying knowledge distillation approaches (Bucila et al., 2006;Hinton et al., 2015) that have proven successful for reducing the size of neural models in other domains to the problem of NMT.”
§ Sequence-Level Knowledge Distillation > Abstract
What the sources say
- Not matchedcites Bucila et al., 2006
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Supportedcites Hinton et al., 2015 source
The Discussion (p. 8) says distillation worked well for transferring knowledge from an ensemble or large model into a smaller model, including in MNIST and acoustic-model experiments. These are successful applications outside NMT, as the citing sentence claims. (LLM, conf=0.94)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Not matched
Every source the paper cites for this claim was checked and supports it.
In the paper
“Training on q(y | x; θ T ) is attractive since it gives more information about other classes for a given data point (e.g.similarity between classes) and has less variance in gradients (Hinton et al., 2015).”
§ Sequence-Level Knowledge Distillation > 2.2 Knowledge Distillation
What the sources say
- Supportedcites Hinton et al., 2015 source
The Introduction (p. 2) says that soft targets from a cumbersome model provide more information per training case and less variance in gradients when they have high entropy. It also explains that probabilities assigned to other classes convey a similarity structure. (LLM, conf=0.93)
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Supported
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Applying weight pruning on top of knowledge distillation results in a student model that has 13× fewer parameters than the original teacher model, with a decrease of 0.4 BLEU.”
§ Sequence-Level Knowledge Distillation > Abstract
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Combining them (Seq-KD + Word-KD) results in further gains for the 2 × 300 and 2 × 100 models (although not for the 2 × 500 model)”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“A beam of size 5 captures 2.8% of the distribution for German → English, and 3.8% for Thai → English.”
§ Sequence-Level Knowledge Distillation > 3.2 Sequence-Level Knowledge Distillation
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We found this to give comparable results to standard knowledge distillation and hence did not pursue this further.”
§ Sequence-Level Knowledge Distillation > 5.3 Further Observations
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“In fact, in experiments we find that with beam of size 1, q(ŷ | s) (on average) accounts for 1.3% of the distribution for German → English, and 2.3% for Thai → English (Table 1: p(t = ŷ)).”
§ Sequence-Level Knowledge Distillation > 3.2 Sequence-Level Knowledge Distillation
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“The perplexity of the baseline 2 × 500 English → German model is 8.2 while the perplexity of the corresponding Seq-KD model is 22.7”
§ Sequence-Level Knowledge Distillation > 7 Conclusion
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We use beam size K = 5 (we did not see improvements with a larger beam).”
§ Sequence-Level Knowledge Distillation > Sequence-Level Knowledge Distillation (Seq-KD)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.
In the paper
“Our student models can even be run efficiently on a standard smartphone.”
§ Sequence-Level Knowledge Distillation > 1 Introduction
What the sources say
- Not matchedcites 1 https://github.com/harvardnlp/nmt-android
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: fails — The paper’s conclusions may not depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. The paper’s conclusions may not depend on it.
- Not matched
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“We find that we can run the student model 10 times faster with greedy decoding than the teacher model with beam search on GPU (1051.3vs 101.9 words/sec), with similar performance.”
§ Sequence-Level Knowledge Distillation > 5.1 Decoding Speed
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
The paper states this without a source or data behind it here, so there is nothing to check it against. The quote shows it is in the paper.
In the paper
“Seq-KD model does significantly better for both greedy (+4.2 BLEU) and beam search (+1.4 BLEU) decoding.”
§ Sequence-Level Knowledge Distillation > 7 Conclusion
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A result the paper reports from its own data — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“While pruning 90% of the weights results in a more appreciable decrease of 1.0 BLEU, the model is drastically smaller with 8m parameters, which is 26× fewer than the original teacher model.”
§ Sequence-Level Knowledge Distillation > 5.2 Weight Pruning
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“indicating that these methods provide orthogonal means of transferring knowledge from the teacher to the student: Word-KD is transferring knowledge at the the local (i.e.word) level while Seq-KD is transferring knowledge at the global (i.e.sequence) level.”
§ Sequence-Level Knowledge Distillation > 5 Results and Discussion
hedgeddirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
In the paper
“Weight pruning for NMT was recently investigated by See et al. (2016), who found that up to 80 -90% of the parameters in a large NMT model can be pruned with little loss in performance.”
§ Sequence-Level Knowledge Distillation > 5.2 Weight Pruning
What the sources say
- Not matchedcites See et al. (2016)
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim.
- Not matched
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Therefore, the fact that sequencelevel knowledge distillation allows for greedy decoding is significant, with practical implications for running NMT systems across various devices.”
§ Sequence-Level Knowledge Distillation > 5.1 Decoding Speed
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
- Rests on an assumption: Run-time complexity for beam search grows linearly with beam size.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The above methods are complementary and can be combined with each other.”
§ Sequence-Level Knowledge Distillation > Sequence-Level Interpolation (Seq-Inter)
plaincorrelational- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“The size of word embeddings have little impact on run-time as the word embedding layer is a simple lookup table that only affects the first layer of the model.”
§ Sequence-Level Knowledge Distillation > 5.2 Weight Pruning
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: unclear — It is unclear why the paper says it holds.
It may state two separate claims as one.
A conclusion the paper draws from its own results — it cites no other source for it, so there is none to check. The quote shows it is in the paper.
In the paper
“Despite its intractability, we posit that this sequence-level objective is worthwhile.It gives the teacher the chance to assign probabilities to complete sequences and therefore transfer a broader range of knowledge.”
§ Sequence-Level Knowledge Distillation > 3.2 Sequence-Level Knowledge Distillation
plaindirect- Asserted: passes — The authors put it forward as true.
- Checkable: fails — It may be too vague to check against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
Probably not a substantial claim. It may be too vague to check against evidence.
The authors state this as an assumption.
The teacher has previously been trained, and we are estimating parameters for the student.
“We generally assume that the teacher has previously been trained, and that we are estimating parameters for the student.”
The authors state this as an assumption.
The teacher distribution parameters remain fixed during student training.
“where θ T parameterizes the teacher distribution and remains fixed.”
The paper relies on this without support.
Run-time complexity for beam search grows linearly with beam size.
“Run-time complexity for beam search grows linearly with beam size.”
The paper relies on this without support.
Combining pruning and knowledge distillation has not been investigated before.
“To our knowledge combining pruning and knowledge distillation has not been investigated before.”
The paper relies on this without support.
The translation domain has generally required the largest capacity deep learning models.
“We have chosen to focus on translation as this domain has generally required the largest capacity deep learning models”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Extracting reasoning matters not because output-only distillation is ineffective, but because reasoning yields a substantially more effective form of capability stealing.”
From Stealing Reasoning Traces from Proprietary LLM APIs (2026) · cites this paper as Kim and Rush, 2016
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“on the 2 × 500 English → German model the word embeddings account for approximately 63% (50m out of 84m) of the parameters.”
§ Sequence-Level Knowledge Distillation > 5.2 Weight Pruning
Reported 63 · recomputed 59.52381
Rated minor: it would not touch a result — a value no result rests on, a p-value that stays on the same side of the significance level, or a conflict that is clerical, resolved elsewhere in the paper, or outside its own findings.
In the paper
“Student Baseline 2 × 500 (Params: 84m) Word-KD Seq-KD Baseline + Seq-Inter Word-KD + Seq-Inter Seq-KD + Seq-Inter Seq-KD + Word-KD Seq-KD + Seq-Inter + Word-KD | 14.7 15.4 18.9 18.5 18.3 18.9 18.7 18.8 | -+0.7 +4.2 +3.6”
§ Sequence-Level Knowledge Distillation > 7 Conclusion
Reported 3.6 · recomputed 3.8
Citations
Cites
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318
Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation
Cho, K., van Merrienboer, B., Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H., & Bengio, Y. (2014) · Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.3115/v1/d14-1179 · Cited by 25,545
BLEU: a method for automatic evaluation of machine translation
Papineni, K., Roukos, S., Ward, T., & Zhu, W.-J. (2001) · Proceedings of the 40th Annual Meeting on Association for Computational Linguistics - ACL '02 · doi:10.3115/1073083.1073135 · Cited by 21,344
Sequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766
Neural Machine Translation by Jointly Learning to Align and Translate
Bahdanau, D., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1409.0473 · Cited by 16,213
Distilling the Knowledge in a Neural Network
Hinton, G., Vinyals, O., & Dean, J. (2015) · arXiv · doi:10.48550/arxiv.1503.02531 · Cited by 14,204
Effective Approaches to Attention-based Neural Machine Translation
Luong, T., Pham, H., & Manning, C. D. (2015) · Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/d15-1166 · Cited by 8,382
Show, Attend and Tell: Neural Image Caption Generation with Visual Attention
Xu, K., Ba, J., Kiros, R., Cho, K., Courville, A., Salakhutdinov, R., Zemel, R., & Bengio, Y. (2015) · arXiv · doi:10.48550/arxiv.1502.03044 · Cited by 8,248
Show and tell: A neural image caption generator
Vinyals, O., Toshev, A., Bengio, S., & Erhan, D. (2015) · 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2015.7298935 · Cited by 6,318
Han, S., Mao, H., & Dally, W. J. (2015) · arXiv · doi:10.48550/arxiv.1510.00149 · Cited by 4,662
Research works printed but not linked: 4
- Cristian Bucila, Rich Caruana, and Alexandru Niculescu-Mizil. 2006. Model Compres- sion. In Proceedings of KDD.
- and Blunsom2013] Nal Kalchbrenner and Phil Blunsom. 2013. Recurrent Continuous Transla- tion Models. In Proceedings of EMNLP.
- Jiwei Li, Michael Galley, Chris Brockett, Jianfeg Gao, and Bill Dolan. 2016. A Diversity- Promoting Objective Function for Neural Conversa- tional Models. In Proceedings of NAACL 2016.
- Mariet and Sra2016] Zelda Mariet and Suvrit Sra. 2016. Diversity Networks. In Proceedings of ICLR.
Cited by
Sorted by citations
Items 701–800 of 841
Orthros: Non-autoregressive End-to-end Speech Translation with Dual-decoderPreprint
Inaguma, H., Higuchi, Y., Duh, K., Kawahara, T., & Watanabe, S. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.13047 · Cited by 1
Model-assisted Learning-based Framework for Sensor Fault-Tolerant Building HVAC Control.Preprint
Xu, S., Fu, Y., Wang, Y., et al. (2021) · arXiv (Cornell University) · Cited by 1
Tokarchuk, E., Rosendahl, J., Wang, W., Petrushkov, P., Lancewicki, T., Khadivi, S., & Ney, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.13097 · Cited by 3
A Cohesive Distillation Architecture for Neural Language ModelsPreprint
Wahle, J. P. (2023) · doi:10.22541/au.167528147.79728645/v1 · Cited by 1
Pretraining Language Models Using Translationese
Doshi, M., Dabre, R., & Bhattacharyya, P. (2024) · Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing · doi:10.18653/v1/2024.emnlp-main.334 · Cited by 1
Vision-Guided Iterative Refinement for Frontend Code Generation
Sansford, H., Law, D. H. C., Liu, W., Tripathi, A., Agarwal, N., & Burg, G. J. J. V. D. (2026) · arXiv · doi:10.48550/arxiv.2604.05839 · Cited by 2
“Subsequent work showed that distillation need not be limited to logits, and can instead operate at the level of entire generated sequences, enabling students to imitate the behavioral outputs of a stronger model in autoregressive settings (Kim & Rush, 2016).”
Stealing Reasoning Traces from Proprietary LLM APIs
Panfilov, A., Schmotz, D., Shumailov, I., Beurer-Kellner, L., Schaeffer, J., Prabhu, A., Geiping, J., & Andriushchenko, M. (2026) · arXiv · doi:10.48550/arxiv.2608.09867
Ding, P., & Zhou, X. (2018) · International Workshop on Semantic Evaluation (SemEval) · doi:10.18653/v1/s18-1174
Neural Machine Translation Models using Binarized Prediction and Error Correction
Oda, Y., Arthur, P., Neubig, G., et al. (2018) · Journal of Natural Language Processing · doi:10.5715/jnlp.25.167
Knowledge Distillation for Machine Translation
Li, Z., Qu, D., Xie, C., et al. (2018) · doi:10.23977/csic.2018.0933
Progressive Gradient Pruning for Classification, Detection and Domain Adaptation
Nguyen-Meidine, L. T., Granger, É., Kiran, M., Pedersoli, M., & Blais-Morin, L.-A. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9412452
Towards Real-time Mispronunciation Detection in Kids' SpeechPreprint
Plantinga, P., & Fosler‐Lussier, E. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.01765
Multi-Agent Cross-Translated Diversification for Unsupervised Machine Translation.Preprint
Nguyen, X.-P., Joty, S., Kui, W., & Aw, A. T. (2020) · arXiv (Cornell University)
Learn to Talk via Proactive Knowledge TransferPreprint
Sun, Q., & Cross, J. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.10077
Research on Mongolian-Chinese Machine Translation Based on Generative Method
Gu, Z., Hou, H., Wu, N., et al. (2020) · IEEE International Conference on Advances in Electrical Engineering and Computer Applications (AEECA) · doi:10.1109/aeeca49918.2020.9213657
Sequence-Level Self-Learning with Multiple Hypotheses
Kumatani, K., Dimitriadis, D., Gaur, Y., et al. (2020) · Interspeech (USB) · doi:10.21437/interspeech.2020-2020
Dual Learning for Speech Processing and Beyond
Qin, T. (2020) · Springer eBooks · doi:10.1007/978-981-15-8884-6_6
Tsinghua University Neural Machine Translation Systems for CCMT 2020
Chen, G., Wang, S., Huang, X., et al. (2020) · Communications in computer and information science · doi:10.1007/978-981-33-6162-1_9
Mode recovery in neural autoregressive sequence modeling
Kulikov, I., Welleck, S., & Cho, K. (2021) · Workshop on Structured Prediction for NLP (SPNLP) · doi:10.18653/v1/2021.spnlp-1.5 · Cited by 1
Investigating Softmax Tempering for Training Neural Machine Translation Models
Dabre, R., & Fujita, A. (2021)
Learning Energy-Based Approximate Inference Networks for Structured Applications in NLPPreprint
Tu, L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2108.12522 · Cited by 1
Unified Distillation: Supervising the Student to Make Fewer Mistakes
Chen, Y., & Rong, P. (2021) · Journal of Physics Conference Series · doi:10.1088/1742-6596/2003/1/012012
Zhang, X., Zhang, X., & Sun, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.12507 · Cited by 2
Reflections of syntactic structures in nonautoregressive language models
HSE, M., & Pletenev, S. (2021) · Computational Linguistics and Intellectual Technologies · doi:10.28995/2075-7182-2021-20-1180-1187
Enhanced encoder for non-autoregressive machine translation
Wang, S., Shi, S., & Huang, H. (2021) · Machine Translation · doi:10.1007/s10590-021-09285-x
Oracle Teacher: Towards Better Knowledge Distillation.Preprint
Yoon, J. W., Kim, H. Y., Lee, H., et al. (2021) · arXiv (Cornell University)
Xu, J., Chen, X., Hu, S., Yu, J., Liu, X., & Meng, H. M. L. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2111.14836
Knowledge Distillation with Metric Learning for Medical Dialogue Generation
Zhu, Q., Wu, P., Tan, Z., et al. (2021) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm52615.2021.9669739
Knowledge Distillation for Translating Erroneous Speech Transcriptions
Fukuda, R., Sudoh, K., & Nakamura, S. (2022) · Journal of Natural Language Processing · doi:10.5715/jnlp.29.344
A Survey of Machine Translation Technologies for Machine Translation on Patent Documents
Imamura, K., Echizen‐ya, H., Ehara, T., Goto, I., Sudoh, K., Sonoh, S., Tsunakawa, T., Nakazawa, T., Ninomiya, T., & Wang, X. (2022) · Journal of Natural Language Processing · doi:10.5715/jnlp.29.925
Random Concatenation: A Simple Data Augmentation Method for Neural Machine Translation
Xiao, N., Zhang, H., Jin, C., & Duan, X. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-17120-8_6
Towards Making the Most of Pre-trained Translation Model for Quality Estimation
Li, C., Di, H., Huang, H., et al. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-18315-7_13
Between Denoising and Translation: Experiments in Text Detoxification
Pletenev, S. (2022) · Computational Linguistics and Intellectual Technologies · doi:10.28995/2075-7182-2022-21-447-455
Lexically Constrained Knowledge Distillation for Neural Machine Translation
Mino, H., Kinugawa, K., Ito, H., et al. (2022) · Journal of Natural Language Processing · doi:10.5715/jnlp.29.1082
Analysis on the number of layers in the transformer-based model for neural machine translation
Li, D., & Luo, Z. (2022) · International Conference on Computer Science and Communication Technology (ICCSCT) · doi:10.1117/12.2661982
Toward cooperative multi-agent video streaming perception
杨, 铮., 董, 亮., & 蔡, 新. (2023) · Scientia Sinica Informationis · doi:10.1360/ssi-2021-0179
Foreground-Specialized Model Imitation for Instance Segmentation
Li, D., Li, W., & Jin, H. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-26293-7_24
Difference-Adaptive Knowledge Distillation for Resource-Constrained Embedded Devices
Xiong, Y., Zhai, W., & Zhu, Z. (2022) · IEEE International Conference on High Performance Computing and Communications; IEEE International Conference on Smart City; IEEE International Conference on Data Science and Systems (HPCC/SmartCity/DSS) · doi:10.1109/hpcc-dss-smartcity-dependsys57074.2022.00297
A Simple Yet Effective Approach to Structured Knowledge Distillation
Lin, W., Li, Y., Liu, L., Shi, S., & Zheng, H.-T. (2023) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49357.2023.10094912
Tan, X. (2023) · Artificial intelligence: foundations, theory, and algorithms/Artificial intelligence: Foundations, theory, and algorithms · doi:10.1007/978-981-99-0827-1_9
Intuitive Access to Smartphone Settings Using Relevance Model Trained by Contrastive Learning
Kim, J.-Y., Lee, K., Shin, H., Lee, H., Kang, S., Choi, B., Shin, D., & Lee, J. (2023) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v37i13.26861
From knowledge augmentation to multi-tasking: towards human-like dialogue systems
Yang, T. H. (2023) · DR-NTU (Nanyang Technological University) · doi:10.32657/10356/168465
Discriminator-Enhanced Knowledge-Distillation Networks
Li, Z., Cao, Z., Li, P., et al. (2023) · Applied Sciences · doi:10.3390/app13148041
On the Interactions of Structural Constraints and Data Resources for Structured Prediction
Zhang, Z., Strubell, E., & Hovy, E. H. (2023) · Workshop on Simple and Efficient Natural Language Processing (SustaiNLP) · doi:10.18653/v1/2023.sustainlp-1.10
Deep Equilibrium Non-Autoregressive Sequence Learning
Zheng, Z., Zhou, Y., & Zhou, H. (2023) · Findings of the Association for Computational Linguistics: ACL (Findings of ACL) · doi:10.18653/v1/2023.findings-acl.747
Hubert, R., Sokolov, A., & Riezler, S. (2023) · International Conference on Spoken Language Translation (IWSLT) · doi:10.18653/v1/2023.iwslt-1.4
Regression-Free Model Updates for Spoken Language Understanding
Caciolai, A., Weber, V., Falke, T., Pedrani, A., & Bernardi, D. (2023) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/2023.acl-industry.52 · Cited by 1
Hybrid Model Compression for BERT-Based Polyphone Disambiguation in Mandarin ChinesePreprint
Zhao, W., wang, Z., & xu, L. (2023) · doi:10.36227/techrxiv.23983191
Hybrid Model Compression for BERT-Based Polyphone Disambiguation in Mandarin ChinesePreprint
Zhao, W., wang, Z., & xu, L. (2023) · doi:10.36227/techrxiv.23983191.v1
A Lightweight Chinese Summarization Generation Method Based on Knowledge Distillation
Liu, S., Xiao, Y., Wang, X., Huang, W., & Deng, Z. (2023) · International Conference on Information Science, Parallel and Distributed Systems (ISPDS) · doi:10.1109/ispds58840.2023.10235472
Higher Target Relevance Parallel Machine Translation with Low-Frequency Word Enhancement
Sun, S., Hou, H., & Wang, Y. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-44198-1_28
Thi, Q.-T. P., Ngo, N.-H., Nguyen, A.-D., et al. (2023) · Communications in computer and information science · doi:10.1007/978-3-031-42430-4_31
Improving neural machine translation: data centric approaches
Nguyen, X. P. (2023) · doi:10.32657/10356/170533
Chen, D.-Y., & Chen, Y.-H. (2023) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc58517.2023.10317576
KD-Boost: Boosting Real-Time Semantic Matching in E-commerce with Knowledge Distillation
Agrawal, S., Sembium, V., & Ankith, M. S. (2023) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2023.emnlp-industry.14
Compute-Efficient Churn Reduction for Conversational Agents
Hidey, C., & Sarthak, S. (2023) · Conference on Empirical Methods in Natural Language Processing: Industry Track (EMNLP) · doi:10.18653/v1/2023.emnlp-industry.28
Compression Models via Meta-Learning and Structured Distillation for Named Entity Recognition
Zhang, Q., Gao, Z., Zhang, M., Duan, J., Wang, H., & He, L. (2023) · International Conference on Asian Language Processing · doi:10.1109/ialp61005.2023.10336991
Distilling ChatGPT for Short Story Generation
Neata, A., & Trăușan-Matu, Ș. (2022) · International Joural of User-System Interaction · doi:10.37789/ijusi.2022.15.3.2
Short-Term Power Load Forecasting based on Distilled Temporal Convolutional Networks
He, Y. T., Luo, F., Yip, L. C., et al. (2024) · IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops) · doi:10.1109/percomworkshops59983.2024.10502465
Advancing Web-Based Visual Question Answering with Efficient Image-Text Alignment
Kilaru, S., Gaur, S., & Rout, S. (2024) · Innovations in Sustainable Technologies and Computing · doi:10.1007/978-981-97-1111-6_28
Riezler, S., & Hagmann, M. (2024) · Synthesis lectures on human language technologies · doi:10.1007/978-3-031-57065-0_2
Biomedical Named Entity Recognition Model Based on Knowledge Distillation
Han, R., Zheng, D., Feng, Y., Li, Y., & Hu, J. J. (2024) · Smart innovation, systems and technologies · doi:10.1007/978-981-97-3980-6_38
LayerGLAT: A Flexible Non-autoregressive Transformer for Single-Pass and Multi-pass Prediction
Li, S., Unanue, I. J., & Piccardi, M. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-70344-7_14
KD-VSUM: A Vision Guided Models for Multimodal Abstractive Summarization with Knowledge Distillation
Zheng, Z., Li, C., Hu, W., & Wang, S. (2024) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn60899.2024.10651189
Summarizing Like Human: Edit-Based Text Summarization with Keywords
Liang, Y., Guo, J., Zhu, Y., & Xu, L. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-72350-6_23
Understanding and Improving Low-Resource Neural Machine Translation with Shallow Features
Sun, Y., Liu, X., Wong, D. F., Lin, Y., Li, B., Zhan, R., Chao, L. S., & Zhang, M. (2024) · Lecture notes in computer science · doi:10.1007/978-981-97-9437-9_18
Harnessing Knowledge Distillation for Enhanced Text-to-Text Translation in Low-Resource Languages
Ahmed, M. O., Ming, Z., & Othmani, A. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-78014-1_22
Moon, S., Yeon, K., Kim, H., et al. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-78447-7_23
Dual Branch Non-Autoregressive Image Captioning
Liu, Y., Yu, H., Li, H., Han, X., & Liu, H. (2024) · Lecture notes in computer science · doi:10.1007/978-3-031-78456-9_21
BeatNet+: Real‑Time Rhythm Analysis for Diverse Music Audio
Heydari, M., & Duan, Z. (2024) · Transactions of the International Society for Music Information Retrieval · doi:10.5334/tismir.198
ERVQ: Leverage Residual Vector Quantization for Speech Emotion Recognition
Xie, J., Xiang, Y., Wang, H., Wu, X., Wu, Z., & Meng, H. M. L. (2024) · International Symposium on Chinese Spoken Language Processing (ISCSLP) · doi:10.1109/iscslp63861.2024.10800636
The Quest of Finding the Antidote to Sparse Double Descent
Quétu, V., & Milovanović, M. (2024) · Communications in computer and information science · doi:10.1007/978-3-031-74640-6_12
Data-Free Functional Projection of Large Language Models onto Social Media Tagging Domain
Mu, W., & Lim, K. H. (2025) · Lecture notes in computer science · doi:10.1007/978-981-96-2054-8_19
Sampling-Based Pruned Knowledge Distillation for Training Lightweight RNN-T
Kim, S., Lee, D., Kang, J. Y., Jeong, M., & Kim, N. S. (2025) · IEEE Signal Processing Letters · doi:10.1109/lsp.2025.3528364
Eghbal-zadeh, H., Wang, Y., Li, R., et al. (2024) · IEEE International Conference on Knowledge Graph (ICKG) · doi:10.1109/ickg63256.2024.00018
Zhang, Y., Zhang, B., Li, Z., Li, M., Cheng, N., Chen, M., Wei, T., Ma, J., Wang, S., & Xiao, J. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10887840 · Cited by 1
Dual Decoder for Fast Inference in Natural Language Generation
Wang, W., Wang, H., Wang, Z., Li, S., & Yu, W. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10889713
Compacting Language Model for Natural Language Understanding on English Datasets
Jeremy, N. H., & Suhartono, D. (2024) · International Conference on Informatics and Computing (ICIC) · doi:10.1109/icic64337.2024.10957202
Compressing Neural Networks on Limited Computing Resources
Lee, S., Lee, D., Hyun, M., Kim, H., & Song, B. C. (2025) · IEEE Access · doi:10.1109/access.2025.3567102
Efficient Algorithms for Leveraging LLMs for Generative and Predictive Recommender Systems
Behdin, K., Dai, Y., Dexter, G., Gupta, A., Mazumder, R., Saha, A., Song, Q., Tang, S., Zhu, S., & Hsu, P.-L. (2025) · International World Wide Web Conference (WWW) · doi:10.1145/3701716.3715868
Fast shallow multi-subnet detector for real-time object detection
Li, Y., Song, M., Hu, K., Chen, S., & Kang, Y. (2025) · Engineering Applications of Artificial Intelligence · doi:10.1016/j.engappai.2025.111090
Fake News Detection on Social Media: Survey
Bibi, A., & Ashraf, H. (2025) · Lecture notes in networks and systems · doi:10.1007/978-981-97-8093-8_29
Rama, S., & Mythili, R. (2025) · International Conference on Data Science and Business Systems (ICDSBS) · doi:10.1109/icdsbs63635.2025.11031972
Improving Neural Machine Translation by Multi-step Teacher-Assistant Knowledge Distillation
Li, F., Ma, H., Chi, C., et al. (2025) · Communications in computer and information science · doi:10.1007/978-981-96-7030-7_14
Enhancing Code Generation for Large Language Models Using Fine-Grained Distillation
Jin, H., Huang, Z., Li, W., Yu, Y., & Wang, C. (2025) · Lecture notes in computer science · doi:10.1007/978-981-95-0014-7_32
Kwarteng, E., Cebe, M., Kwarteng, J. L., & Bansal, N. (2025) · International Journal of Information Security · doi:10.1007/s10207-025-01094-0 · Cited by 1
Enhancing Low-Resource Machine Transliteration Through Three-Way Transfer Learning
Thamilselvan, R., Kalpana, T., Natesan, P., Prasath, S., & Balaji, B. S. (2025) · International Conference on Computing Technologies (ICOCT) · doi:10.1109/icoct64433.2025.11118514
ADDL: Adaptive Dynamic Distillation for Large Language Models
Zhang, R., Zhou, K., Zhang, H., et al. (2025) · International Conference on Computer Science and Blockchain (CCSB) · doi:10.1109/ccsb66722.2025.11154259
Translating Images to Road Network: A Sequence-to-Sequence Perspective
Lu, J., Nie, M., Zhang, B., Peng, R., Cai, X., Xu, H., Wen, F., Zhang, W., & Zhang, L. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3612940
Semantic Communication with Transformer and Knowledge Distillation for Language Translation
Nguyen, H. T., Nguyen, T. A., Kieu, T., Nguyen, T.-V., & Vo, P. L. (2025) · Communications in computer and information science · doi:10.1007/978-3-032-10209-6_8
Alleviating Directional Bias in Non-Autoregressive Transformers
Wang, S., Wan, Y., Wong, D. F., Zhang, S., Lin, Y., & Chao, L. S. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228036
Shared Latent Space by Both Languages for Non-Autoregressive Neural Machine Translation
Heo, D., & Choi, H. (2025) · International Joint Conference on Neural Networks (IJCNN) · doi:10.1109/ijcnn64981.2025.11228007
Optimizing T5 for Lightweight Tibetan-English TranslationPreprint
Moore, J., & Lauren, P. (2025) · Research Square · doi:10.21203/rs.3.rs-7409829/v1
Lightweight Zero-Shot Keyword Spotting via Multi-Granular Knowledge Distillation
Sun, Y.-T., Li, L.-Y., Lo, T.-H., et al. (2025) · Asia-Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC) · doi:10.1109/apsipaasc65261.2025.11249226
A High-Performance Knowledge Distillation Framework Based on Temperature Decoupling
Qin, H., Liu, H., & Zhang, L. (2025) · International Conference on Intelligent Technology and Embedded Systems (ICITES) · doi:10.1109/icites66466.2025.11274290
Evolving Knowledge Distillation for Lightweight Neural Machine Translation
Zhang, X., Zhang, H., & Huang, X. (2025) · Proceedings - International Conference on Tools with Artificial Intelligence, TAI · doi:10.1109/ictai66417.2025.00092
Zhao, Y., Zhang, Y., Xiang, L., Zhang, J., Zhou, Y., & Zong, C. (2025) · IEEE Transactions on Audio Speech and Language Processing · doi:10.1109/taslpro.2025.3623904
Distillation Progressive Teacher-student Architecture to Improve Abstractive SummarizationPreprint
Wang, G., Wang, D., Liu, S., Wang, K., Liu, X., Qiang, J., & Liu, J. (2025) · SSRN Electronic Journal · doi:10.2139/ssrn.5731335
Oh, S., Zhao, J., Russo, C., Bolmer, M., Jeong, J., Fan, J.-X., Cao, Y.-S., Wang, W.-L., Andrus, N., & McCrickard, S. (2025) · Frontiers in Education Conference · doi:10.1109/fie63693.2025.11328660
Metadata
- Authors
- Kim, Y., & Rush, A. M.
- Affiliations
- Harvard University (US)
- Venue
- Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing
- Published
- 2016
- Pages
- 1317-1327
- OpenAlex
- W2963736842
- Semantic Scholar
- 57a10537978600fd33dcdd48922c791609a4851a
- Cited by
- 836
- References
- 53
- Citations by year
- 2026: 36 · 2025: 59 · 2024: 52 · 2023: 137 · 2022: 115 · 2021: 157 · 2020: 143 · 2019: 75 · 2018: 42 · 2017: 16 · 2016: 4
- FWCI
- 24.59
- Type
- conference paper
- Language
- en
- Open access
- hybrid
- doi.org
- Licence
- cc-by
- Locations
- Conference on Empirical Methods in Natural Language Processing (EMNLP) (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Keywords
- knowledge distillation, neural machine translation, model compression, weight pruning, teacher-student models, parameter reduction
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
A language model drew the claims from the paper on 2026-10-23, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-23 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
