Diffusion language models (DLMs) can write several tokens per forward pass. Open-weight DLMs commit, in any order, every masked position whose confidence passes a threshold. If none does, a fallback commits the most confident one, and the commits are never revised. Wide windows let the decoder commit far ahead of the text, which should speed up decoding but often costs accuracy. It is unclear where the model is confident and which commits cause errors. We trace every commit of four off-the-shelf DLMs on math, code, translation and summarization, and rerun failing answers with one step changed. Most tokens become confident only near the frontier, the leftmost masked position, so committing only there, still several tokens per step, is about as fast as committing anywhere. The accuracy loss comes from committing tokens before the text to their left, and committing one token per step does not reduce it. The worst case is an early end-of-sequence token that cuts the answer short. Fallback commits are 11-35% of commits but start 61-72% of the failures that left-to-right decoding avoids. Changing only the step where such a failure starts repairs it more often than changing a similar earlier step. Committing only at the frontier recovers most of the lost accuracy on all four models. Our code is available at https://github.com/rmakarovv/fast-at-the-frontier.
Figures & tables
Figure 1: One real 32-token block (NLD-3B, GSM8K, vanilla ); each box shows a token and its commit step. Blue: commits at the frontier; orange: ahead commits, made while a position to their left was still masked (“miles” at step 0, before “15”); dashed: fallback commits.
Figure 2
Policy
Tokens committed per step
vanilla
One or more. Commit all tokens with confidence ≥0.9 , otherwise the most confident one.
one-per-step
One. Commit the most confident masked position.
left-to-right
One. Commit the frontier position (left-to-right decoding).
frontier-only
One or more. Commit the confident positions in a row, starting at the frontier. If none exists, commit the frontier position (PARD; Hou and Kwok, 2026 ).
Variants of vanilla : one or more tokens per step
frontier-fallback
Use the frontier position as the fallback (Fork-dLLM; Frković et al., 2026 ).
Table 1: Decoding policies. All policies use the same model, prompt, and confidence threshold ( 0.9 ). Commits are restricted to the decoding window. The first four vary position order and token count. The remaining policies modify vanilla .
Figure 2: Most tokens become confident only near the frontier, so wider windows add little speed. (a) How far ahead of the frontier answer positions first pass the threshold in left-to-right decodes (GSM8K; NLD B=128 , LLaDA B=256 prompt A, Fast-dLLM v2 B=128 ), in % of answer positions per distance. (b) Tokens per step by B (NLD-3B, vanilla , blocks inside the answer). (c) TPF of frontier-only over TPF of vanilla at the same window, against vanilla ’s ahead share; one marker per model, task and window (34 in all). The two high points are settings where vanilla cuts answers short.
confidence order
frontier order
effect (points)
EOS only at the frontier
Model
B
several vanilla
one one-per-step
one left-to-right
several frontier-only
order
paral- lelism
several eos-at-frontier
one one-per-step
order
NLD-3B
128
79.1
79.7
88.0
87.6
+8.4
−0.5
86.2
87.3
+1.1
NLD-3B
256
54.5
55.0
88.0
88.1
+33.3
−0.2
87.2
88.0
+0.5
NLD-8B
128
87.3
86.7
90.2
90.1
+3.2
+0.2
91.0
91.0
−0.8
NLD-8B
256
73.8
73.3
88.9
88.8
+15.3
+0.2
90.4
90.2
−1.4
LLaDA, A
256
50.6
49.9
71.9
71.6
+21.5
+0.2
54.0
53.9
+17.8
Table 2: Frontier order beats confidence order at every wide window; committing one token per step does not help. Allowing EOS only at the frontier removes the gap on NLD, but not on LLaDA or Fast-dLLM v2. GSM8K accuracy (%) at each model’s wide window. Order : frontier minus confidence order, averaged over one and several tokens per step, 21[(frontier-only−vanilla)+(left-to-right−one-per-step)] ; parallelism : several minus one token per step, averaged over both orders. EOS only at the frontier : the confidence-order policies rerun with this restriction and the order effect recomputed; near zero means that early EOS tokens caused the order effect. Intervals in Section C.1 .
Policy
NLD 3B
NLD 8B
LLaDA A
LLaDA B
Fast- dLLM v2
vanilla
79.1
87.3
50.6
35.5
79.4
frontier-fallback
85.7
89.6
67.9
75.4
82.6
frontier-gating
87.3
90.9
72.6
80.7
80.8
entropy-budget
87.1
91.4
73.2
80.8
80.3
eos-at-frontier
86.2
91.0
54.0
57.4
79.5
frontier-only
87.6
90.1
71.6
82.3
85.4
Table 3: Frontier policies recover most of the loss; only frontier-only does so on all four models. GSM8K accuracy (%) at the wide window (NLD B=128 ; LLaDA B=256 , prompts A and B; Fast-dLLM v2 B=128 ). Last row: vanilla at each model’s default window (NLD and LLaDA B=32 ; Fast-dLLM v2, 8-token sub-blocks).
Fallback share (%)
Rate ratio
Model
commits
harmful roots
fallback / confident
ahead / frontier
NLD-3B
14
61
20.2
9.6–14.6
NLD-8B
35
72
31.5
9.0–14.3
Fast-dLLM v2
11
70
19.1
7.8–12.3
LLaDA
17
68
10.3
1.3–1.9
Table 4: Most harmful departures start at fallback commits. Wide windows (NLD B=128 , four tasks; Fast-dLLM v2 B=128 ; LLaDA B=256 , prompts A and B). Fallback share : of all commits, and of the roots of harmful departures. Rate ratios per commit: harmful roots at fallback vs confident commits, and departures (harmful and helpful) at distance ≥1 vs 0 at the same confidence (range over distance bins; the reference commits the frontier, which lowers the frontier rate).
Appendix figures & tables16 assets
Supplementary material from the paper’s appendix.
Appendix
Task
Source and split
Size
Metric
Models
Answer length
GSM8K ( Cobbe et al., 2021 )
openai/gsm8k, test
1319
accuracy
all four
179 tokens
MATH-500 ( Hendrycks et al., 2021 ; Lightman et al., 2024 )
HuggingFaceH4/MATH-500
500
accuracy
NLD, Fast-dLLM v2
591 tokens
HumanEval ( Chen et al., 2021 )
EvalPlus ( Liu et al., 2023 ) , base tests
164
pass@1
all four
552 tokens
MBPP ( Austin et al., 2021b )
EvalPlus ( Liu et al., 2023 ) , base tests
378
pass@1
all four
461 tokens
WMT16 de → en ( Bojar et al., 2016 )
wmt/wmt16, test, 500 random sentences
500
chrF ( Popović, 2015 )
NLD-3B, LLaDA
18 words
CNN/DailyMail ( Hermann et al., 2015 ; See et al., 2017 )
abisee/cnn_dailymail 3.0.0, test, 500 random articles
500
ROUGE-L ( Lin, 2004 )
NLD-3B, LLaDA
72 words
Appendix
Table 5: Tasks. Size: number of problems. Answer length: mean over the vanilla answers of NLD-3B at B=32 (tokens for math and code, words for translation and summarization). The CD4LM experiment ( Section 5.2 ) uses that work’s own 440-problem GSM8K subset.
Figure 3: A larger block mainly adds positions. Mean commit step by position for B=8 –128 (NLD-3B, GSM8K, vanilla ); the curves nearly coincide.
Model
Task
Window
Ahead
vanilla
one-per-step
frontier-only
TPF vanilla
TPF frontier-only
NLD-3B
GSM8K
4
6
88.6
88.9
–
1.51
–
NLD-3B
GSM8K
8
13
89.1
89.2
–
2.06
–
NLD-3B
GSM8K
16
23
87.9
88.8
–
2.46
–
NLD-3B
GSM8K
32
32
87.6
88.3
88.1
2.75
2.61
NLD-3B
GSM8K
64
41
87.0
88.9
88.4
2.72
2.74
NLD-3B
GSM8K
128
49
79.1
79.7
87.6
2.27
2.77
Appendix
Table 6: All model × task × window settings (greedy). Ahead: ahead commits among the commits of vanilla (%). Accuracy (%) of vanilla , one-per-step and frontier-only ; TPF: tokens per forward. Window: B for NLD and LLaDA (256 = one block); for Fast-dLLM v2, the sub-block for 4–16, the whole 32-token training block at 32, and B for 64 and 128. A dash marks a setting we did not run.
Decoder
batch 1
batch 8
batch 32
self-speculation
4.72
3.42
1.93
+ LoRA drafter
4.83
3.60
2.00
threshold decoding
vanilla , B=32
2.95
1.57
0.63
frontier-only , B=32
2.85
1.54
0.62
vanilla , B=128
3.12
0.88
0.25
Appendix
Table 7: NLD-3B throughput relative to AR decoding on the same GPU. Steady-state throughput over the first 512 tokens of 64 long MATH-500 answers (A100-40GB, eager PyTorch). The self-speculation and B=32 rows come from one benchmark. The larger-block rows come from a second benchmark of two jobs run in opposite order, which agree within 3%.
Task
vanilla
frontier-only
frontier-fallback
frontier-gating
entropy-budget
eos-at-frontier
vanilla , B=32
GSM8K
79.1
87.6
85.7
87.3
87.1
86.2
87.6
MATH-500
69.6
73.2
68.8
71.8
72.2
70.8
74.4
HumanEval
64.6
81.1
73.2
70.7
71.3
70.1
73.2
MBPP
63.5
72.5
65.3
72.0
71.7
69.8
71.2
Appendix
Table 8: NLD-3B accuracy (%) by task and policy. Policies at B=128 , and vanilla at B=32 for comparison; Table 9 gives one-per-step and left-to-right .
Model, window
task
vanilla
one-per-step
left-to-right
frontier-only
order
parallelism
interaction
NLD-3B, B=128
GSM8K
79.1
79.7
88.0
87.6
+ 8.4 ± 1.7
− 0.5 ± 0.8
+ 0.2 ± 1.5
NLD-3B, B=128
MATH-500
69.6
73.8
76.6
73.2
+ 3.2 ± 2.9
− 3.8 ± 2.0
+ 0.8 ± 3.3
NLD-3B, B=128
HumanEval
64.6
70.1
80.5
81.1
+ 13.4 ± 6.3
− 2.4 ± 2.7
+ 6.1 ± 5.2
NLD-3B, B=128
MBPP
63.5
64.3
73.8
72.5
+ 9.3 ± 4.4
− 1.1 ± 1.9
− 0.5 ± 3.6
NLD-3B, B=256
GSM8K
54.5
55.0
88.0
88.1
+ 33.3 ± 2.6
− 0.2 ± 1.0
+ 0.6 ± 2.0
NLD-8B, B=128
GSM8K
87.3
86.7
90.2
90.1
+ 3.2 ± 1.5
+ 0.2 ± 0.9
− 0.6 ± 1.8
Appendix
Table 9: Order × parallelism in every setting with all four policies (accuracy, %; effects in points, ± half the width of the 95% problem-bootstrap interval; Section 5.1 ). vanilla : confidence order, several tokens per step; one-per-step : confidence order, one; left-to-right : frontier order, one; frontier-only : frontier order, several.
accuracy, EOS only at the frontier
order effect
Model
window
task
vanilla
one-per-step
left-to-right
frontier-only
without
with
par.
int.
kept
NLD-3B
B= 128
GSM8K
86.2
87.3
88.0
87.6
+8.4 ± 1.7
+1.1 ± 1.3
− 0.8 ± 0.6
+0.6 ± 1.1
13%
NLD-3B
B= 128
MATH-500
70.8
74.2
76.6
73.2
+3.2 ± 2.9
+2.4 ± 2.8
− 3.4 ± 1.8
+0.0 ± 2.7
75%
NLD-3B
B= 128
HumanEval
70.1
73.2
80.5
81.1
+13.4 ± 6.3
+9.1 ± 5.3
− 1.2 ± 2.4
+3.7 ± 5.2
68%
NLD-3B
B= 128
MBPP
69.8
72.2
73.8
72.5
+9.3 ± 4.4
+2.1 ± 4.0
− 1.9 ± 1.7
+1.1 ± 3.0
23%
NLD-3B
B= 256
GSM8K
87.2
88.0
88.0
88.1
+33.3 ± 2.6
+0.5 ± 1.3
− 0.4 ± 0.6
+0.9 ± 1.1
1%
Appendix
Table 10: Order × parallelism with EOS allowed only at the frontier in every cell: vanilla becomes eos-at-frontier , and one-per-step is rerun with the same restriction; left-to-right and frontier-only commit only at the frontier, where the restriction changes nothing. Accuracy (%) and effects (points, 95% problem bootstrap). Par. and int.: parallelism and interaction effects. Kept: the order effect with the restriction as a share of the order effect without it; it is negative where the effect changes sign.
Setting
0.7
0.8
0.9
0.95
NLD-3B, B=128
73.9 / 85.5 (+11.6)
77.3 / 86.9 (+9.6)
79.1 / 87.6 (+8.5)
79.5 / 87.9 (+8.4)
TPF
4.0 / 3.8
3.0 / 3.2
2.3 / 2.8
1.8 / 2.5
NLD-3B, B=128 , MATH-500
61.0 / 67.2 (+6.2)
67.0 / 70.8 (+3.8)
69.6 / 73.2 (+3.6)
70.2 / 75.0 (+4.8)
TPF
5.2 / 4.4
4.4 / 4.1
3.5 / 3.4
3.1 / 3.0
NLD-3B, B=32
81.0 / 85.4 (+4.4)
84.7 / 87.1 (+2.4)
87.6 / 88.1 (+0.5)
88.6 / 88.2 ( −0.4 )
TPF
4.0 / 3.4
3.3 / 3.1
2.7 / 2.6
2.4 / 2.3
Appendix
Table 11: Threshold sweep. Accuracy (%) of vanilla / frontier-only with the gap ( frontier-only minus vanilla , points), and below it the TPF of vanilla / frontier-only . NLD-3B; LLaDA B=256 (prompt B); Fast-dLLM v2 B=128 ; GSM8K unless stated.
Figure 4: Early EOS commits cause most of the wide-window loss on NLD and most of LLaDA’s reported collapse. (a) NLD: accuracy of the answers whose EOS token the fallback committed early, and of the same problems under eos-at-frontier . (b) LLaDA-8B, GSM8K, in the released CD4LM evaluation pipeline (decoder, prompt and grader; 440 problems). Each of the three middle bars changes one setting of the released one-block setup; committing the leftmost masked position first removes the drop.
Model, task
k=0
1
2
4
8
16
32
∞
NLD-3B, GSM8K
acc.
87.6
87.9
87.9
87.8
87.6
87.1
86.4
79.1
Δ
0
+0.4
+0.4
+0.2
+0.1
− 0.5
− 1.1
− 8.5
± 1.3
± 1.4
± 1.5
± 1.4
± 1.4
± 1.5
± 2.0
TPF
2.77
2.81
2.82
2.85
2.87
2.86
2.80
2.27
NLD-3B, MATH-500
acc.
73.2
73.2
73.6
72.4
72.0
69.8
69.2
69.6
Δ
0
+0.0
+0.4
− 0.8
− 1.2
− 3.4
− 4.0
− 3.6
Appendix
Table 12: Distance cap: accuracy (%), its difference to k=0 ( Δ , points, ± half the width of its paired 95% problem-bootstrap interval) and tokens per forward when confident commits and the fallback may reach at most k positions past the frontier; k=0 is frontier-only , k=∞ vanilla .
Figure 5: Fallback commits start most departures, and on the block-diffusion models, distance from the frontier adds risk at the same confidence. (a) Departures (harmful or helpful) started per confident commit, by distance to the frontier, relative to a frontier commit at the same confidence (Poisson model with answer position, EOS token and task; log scale; 95% intervals). (b) Shares of all commits, of harmful roots and of helpful roots by kind, at wide windows (NLD B=128 , Fast-dLLM v2 B=128 , LLaDA B=256 ). Reference: each model’s own left-to-right decode; commits counted up to the root’s step.
Term
NLD-3B
NLD-8B
Fast-dLLM v2
LLaDA
all departures
distance 1
10.9 −1.5+2.0
9.0 −1.0+1.2
8.7 −1.4+1.6
1.6 −0.3+0.4
distance 2–3
12.7 −1.9+2.2
10.7 −1.2+1.7
11.9 −1.9+1.8
1.9 −0.3+0.2
distance 4–7
14.6 −2.2+2.9
11.9 −1.5+2.0
12.3 −1.9+2.4
1.3 ± 0.2
distance 8–15
10.2 −2.0+2.4
14.3 −2.1+2.4
10.8 ± 2.2
1.4 −0.4+0.3
distance ≥16
9.6 −1.9+2.6
12.0 −2.3+2.7
7.8 −1.5+2.0
1.5 −0.4+0.3
Appendix
Table 13: Poisson model (reference: each model’s own left-to-right decode with the same prompt): rate ratios of roots per commit against a frontier commit at the same confidence (nine bins), answer position, EOS token and task; wide windows (NLD B=128 , four tasks; Fast-dLLM v2 B=128 , GSM8K and MATH-500; LLaDA B=256 , prompts A and B); 95% intervals from 500 problem resamples within tasks, given as the distance to their upper (superscript) and lower (subscript) end, or ± when the two are equal. Per-task values in the order GSM8K, MATH-500, HumanEval, MBPP (LLaDA: prompts A, B).
roots
fallback share
fallback share of roots
fallback / confident per commit
fallback share
Model
harmful
helpful
of commits
harmful
helpful
harmful
helpful
of the net gap
NLD-3B
299
81
14%
61% ± 6
74% ± 9
20.2 −5.7+7.3
19.1 −6.7+13.4
56% ± 8
NLD-8B
163
105
35%
72% ± 7
90% ± 5
31.5 −14.0+52.4
19.9 −8.1+27.1
38% −69+23
Fast-dLLM v2
239
75
11%
70% ± 6
73% ± 10
19.1 −4.3+6.8
22.3 −8.9+14.2
69% −10+9
LLaDA
894
125
17%
68% ± 3
69% ± 8
10.3 −1.4+1.5
10.7 −3.1+5.7
68% ± 4
Appendix
Table 14: Harmful and helpful roots at wide windows (the runs and reference of Table 4 , commits counted up to the root). Helpful: the vanilla answer is right and the reference answer wrong. Shares ± half the width of the 95% interval; rate ratios (roots per commit, fallback vs confident) and the fallback share of the net accuracy gap to the reference (harmful minus helpful roots; a decomposition of the gap, not a causal estimate) with the distance to the upper (superscript) and lower (subscript) end of their 95% interval, or ± when the two are equal; 500 problem resamples within tasks.
Model
Task
B
n
Root
Ctrl.
Root − ctrl.
NLD-3B
GSM8K
128
77
44.2
29.2
+14.9±9.0
NLD-3B
GSM8K
256
325
29.5
22.3
+7.2±3.2
NLD-3B
MATH
128
33
24.2
6.3
+17.9±14.9
Fast-dLLM v2
GSM8K
128
114
47.4
8.2
+39.2±8.7
Fast-dLLM v2
MATH
128
73
28.8
10.6
+18.2±8.7
Appendix
Table 15: At the root’s step, committing only at the frontier repairs more failing answers than at matched control steps. Repaired share (%) of failing answers when the root’s step or a matched earlier control step (ctrl.) is changed, and their paired difference (points, ± half the width of the 95% problem-bootstrap interval); n : failing answers replayed. Table 16 gives sampled steps and helpful roots.
sampled steps: repaired
helpful roots: broken
Setting
n
fallback
confident
n
root
sampled
NLD-3B GSM8K, B=128
440
29.6
17.3
19
36.8
7.9
NLD-3B MATH-500, B=128
172
4.2
3.0
14
35.7
17.9
NLD-3B GSM8K, B=256
1608
19.7
18.0
21
23.8
2.4
Fast-dLLM v2 GSM8K, B=128
592
16.6
9.6
50
34.0
5.0
Fast-dLLM v2 MATH-500, B=128
364
8.1
6.8
22
31.8
15.9
Appendix
Table 16: Replays at sampled steps and at helpful roots ( Section D.2 ). Sampled: steps drawn uniformly from all steps of the failing answer that commit ahead of the frontier, without using the root; repaired share (%) at fallback and confident steps. Helpful: answers that vanilla gets right and the reference gets wrong; share (%) broken by the same change at the root’s step and, averaged, at two sampled steps of the same answers. n : sampled steps (four per failing answer) and helpful departures.
Frozen tokens
acc.
vs default
share of loss
none (default)
94.3
–
–
frontier tokens
93.8
−0.4±0.4
15%
step-0 tokens
93.6
−0.7±0.4
24%
as many random (control)
92.6
−1.6±0.5
60%
ahead tokens
92.0
−2.2±0.5
81%
all accepted tokens
91.5
−2.7±0.5
100%
Appendix
Table 17: DiffusionGemma-26B with frozen tokens. Accepted tokens are frozen at their first acceptance, with the same acceptance budget. Three seeds averaged within each problem; four tasks pooled (2361 problems); HumanEval graded leniently. Differences to the default sampler ± half the width of the 95% interval from resampling problems. Share of loss: each drop as a share of the 2.7-point drop from freezing all accepted tokens. The control freezes as many tokens at the same steps, at random positions.
Diffusion language model decoding generates text through iterative parallel refinement, balancing fewer denoising steps with accuracy, coherence, and control.
Diffusion language models generate multiple token positions in parallel, but few-step decoding makes parallel commitment difficult: tokens that appear reliable individually may be jointly inconsistent, while premature commitments can prevent later context-dependent refinement. In this paper, we introduce SetCommit, a training-free method for coordinated parallel token commitment. Instead of selecting tokens independently by confidence, SetCommit selects a commitment set that balances token reliability against local joint commitment risk, preserving conditioning opportunities for subsequent denoising steps. It further performs budget-neutral local exchanges, replacing vulnerable earlier commitments when newly revealed context supports more reliable alternatives. SetCommit requires no retraining or architectural modification and is compatible with existing diffusion decoding strategies. Experiments with LLaDA-8B and Dream-7B on GSM8K, MATH-500, MBPP, and HumanEval show accuracy/pass@1 gains of up to 7.4 percentage points over native decoding at 32 model evaluations. At common quality targets, geometric-mean incremental speedups range from 1.44 to 1.74×. Code is available at https://github.com/Kissmetothemoon/SetCommit.
Yuannuo Feng, Yubing Ye, Wenshuai Yao +4
Beihang University · The University of Hong Kong · Peking University
Parallel decoding speeds up masked diffusion language models by writing several tokens per forward pass. Published commit rules choose these tokens by testing the predicted distributions. Under sampling the written tokens are drawn, and these tests do not see them. We find that answer quality follows the reliability of the tokens actually written. An answer's log probability splits into this reliability and the dependence within each step. We propose PACT, a training-free commit rule that draws the tokens first, certifies the set it will write and spaces the positions of each step apart. This certificate holds at every temperature, and we show that no rule with the same guarantee commits more positions. Experiments use four models of the LLaDA and Dream families and HumanEval, MBPP, GSM8K, MATH500 and IFEval benchmarks. PACT reaches the quality of published commit rules with fewer forward passes. Spacing removes positions from each step, yet it usually lowers the total cost because later steps then write more tokens. At the same speed, published decoders also write a word twice in a row far more often than PACT. These results suggest that commit rules should certify the tokens they write. Code is available at https://github.com/LiZaiyuan0619/PACT.
Yanchao Li, Yusu Zhang, Jiaqing Xie +5
Nanjing University · Shanghai Artificial Intelligence Laboratory
Diffusion language models predict tokens in parallel but still require many iterative forward passes. Adaptive sampling accelerates local token commitment, while recent global stopping methods seek to end generation early. Our trajectory analysis shows that short answers can stabilize early in decoding, whereas reasoning outputs often require more steps to settle. A fixed progress threshold therefore struggles to avoid both redundant computation on short-answer tasks and premature termination on reasoning tasks. We introduce C4, a training-free framework that combines local decoding acceleration with adaptive global early exit. Confidence-Verified Early Exit (CVEE) determines when to end generation, skipping the remaining steps once its stopping criterion is met. It combines confidence and temporal agreement to adapt stopping to the observed revision history. Commit-Core-Then-Confirm (CCTC) accelerates token commitment while generation continues, reducing the decoding steps needed to complete each block. It extends confidence-threshold parallel decoding with a boundary-anchored core and one-step confirmation of deferred predictions, constraining additional commitments to mitigate premature freezing under unresolved context. With CVEE calibrated once on LLaDA, the same numerical settings transfer across all 12 zero-shot tasks and to Dream without recalibration. Under this configuration, C4 reduces decoding steps by 64-95% and achieves measured decoding-loop speedups of 2.6x-18.6x, with task performance close to full decoding. Component analyses show complementary contributions: early exit supplies most savings on short-answer tasks, while local acceleration contributes more on reasoning and code generation.
Chia-Ming Lee, Shao-Kai Liu, Ming-Ching Chang +3
National Yang Ming Chiao Tung University · University at Albany, SUNY