Claude for Research

127

Experiments Logged

14

Evals Tested

3

Domains Explored

+0.9%

Avg Score Gain / Run

Program DatabaseLive · 22 entries

13 kept6 reverted3 flagged

	ID	Hypothesis	Domain	Score	Δ	Status	Time
	EXP-047	LR schedule: cosine annealing with T_max=500	Transformer opt.	0.742	+1.4%	KEPT	2m ago
	╰	Dropout rate increase 0.2→0.3 in FFN layers	Transformer opt.	0.728	-0.3%	REVERTED	8m ago
	╰	Weight decay 0.01 → 0.001	Transformer opt.	0.731	+0.9%	KEPT	14m ago
	╰	Gradient clipping max_norm 1.0 → 0.5	Transformer opt.	0.724	+0.4%	KEPT	19m ago
	╰	WordPiece tokenization for code-specific tokens	Transformer opt.	0.718	+3.4% BLEU	FLAGGED	25m ago
	EXP-046	Graph conv pooling: mean → attention-weighted sum	Drug discovery	0.902	+1.2%	KEPT	28m ago
	╰	SMILES dropout augmentation rate 0.1 → 0.2	Drug discovery	0.896	+0.7%	KEPT	34m ago
	╰	Node feature normalization: batch norm → layer norm	Drug discovery	0.888	-0.5%	REVERTED	41m ago
	EXP-047	Label smoothing ε=0.1 applied to SFT objective	Transformer opt.	0.713	+0.1%	KEPT	47m ago
	╰	Batch size linear ramp 32 → 128 over first 1k steps	Transformer opt.	0.704	+0.5%	KEPT	53m ago
	EXP-015	TD3 policy update delay: 2 → 4 steps	RL robotics	0.661	+2.1%	KEPT	58m ago
	╰	Exploration noise std 0.1 → 0.2 with decay schedule	RL robotics	0.653	+0.9%	KEPT	1h 5m ago
	╰	Critic network: 2 layers → 3 layers, hidden 256	RL robotics	0.648	-0.6%	REVERTED	1h 12m ago
	╰	Polyak averaging coefficient 0.995 → 0.999	RL robotics	0.655	+1.1%	FLAGGED	1h 19m ago
	EXP-046	3D conformer features concatenated to fingerprint	Drug discovery	0.881	+2.3%	FLAGGED	1h 27m ago
	╰	Flash Attention 2 kernel — latency vs throughput	Transformer opt.	0.700	−1ms	KEPT	1h 44m ago
	╰	Layer norm epsilon: 1e-5 → 1e-8	Transformer opt.	0.698	+0.2%	KEPT	1h 59m ago
	╰	Increased attention dropout 0.0 → 0.1	Transformer opt.	0.691	-1.1%	REVERTED	2h 14m ago
	EXP-091	Multi-task loss weighting: inverse class frequency	Drug discovery	0.874	+1.8%	KEPT	2h 38m ago
	╰	Dropout 0.1 → 0.3 in message-passing layers	Drug discovery	0.857	-0.4%	REVERTED	2h 52m ago
	EXP-013	Reward normalization with running mean/std	RL robotics	0.633	+0.8%	KEPT	3h 10m ago
	╰	Policy network hidden size 256 → 512	RL robotics	0.622	-0.5%	REVERTED	3h 25m ago

Eval Reliability

Predictive power across experiments

Val Loss

r=0.9194%

47 experiments

Pass@1

r=0.8588%

47 experiments

BLEU-4

r=0.6972%

47 experiments

Inference Latency

r=0.5861%

47 experiments